# Multica

Podium publishes this catalog and sells an agent development environment. Podium chooses the products, criteria, sources and wording, so treat these pages as vendor-authored research. Check important claims against the linked primary sources. The same evidence rules apply to Podium and its competitors, but they do not remove this conflict of interest.

Page: https://adecomparison.com/ade/multica/
Website: https://multica.ai/
Research: covered

Method: https://adecomparison.com/method/

Rank by must-haves met, then preferences, then name. Read each finding's checked date. Findings stay published after their scheduled recheck. Checks over six months old may be outdated.

## multica:ACC-01

Finding: Web app, Desktop app (macOS, Windows, Linux) and a CLI exist; the iOS client is self-built from source, with no App Store or Android client. Release stage is not labelled and versions are pre-1.0. The browser issue view includes a composer layout specifically for phone-sized viewports.

Verification: sourced
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Multica Cloud/web/desktop; Self-built iOS client
Version: Live docs reread 2026-11-07 through v0.8.1; source at v0.6.1 commit 8db6cfe

Only the named atomic answers are established by this source pass. Remaining atomic answers retain their documented gaps.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/mobile-app
Locator: Intro; Prerequisites; 7-day limit of free signing
Finding: Current mobile guide says source-built iPhone client, no App Store and no Android, seven-day free signing, server-owned workspace data. Native iPad support is separately documented in release notes; no device test.
Retrieved: 2026-11-07

Source: https://multica.ai/changelog
Locator: v0.8.1 (November 6, 2026), v0.8.0 (November 3) and historical release entries
Finding: Current changelog lists v0.8.1 dated November 6, 2026, after v0.8.0 November 3. Skills revision history, unread filters and Qoder updates are released, not future promises.
Retrieved: 2026-11-07

Source: https://multica.ai/download
Locator: Choose your platform; Prefer the CLI?; Cloud runtime (waitlist); Current version
Finding: Downloads show macOS Apple Silicon/Intel, Windows x64/ARM64 and Linux x64/ARM64 installers; current v0.8.1. Vendor-hosted runtime remains not live, waitlist-only. It is separate from connecting Qoder Cloud Agents.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/desktop-app
Locator: Installation; Built-in daemon; Updating Desktop; Connecting to a self-hosted instance
Finding: Desktop Mac/Windows/Linux shares server data with web, manages separate daemon profiles, auto-updates with OS/architecture feeds and supports self-hosted apiUrl configuration. Terminal issue CLI installed separately.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/cloud-quickstart
Locator: Steps 1-5; Common problems
Finding: Web/Desktop sign-in uses email code or Google. Desktop connects a computer automatically; other computers install CLI and run setup. At least one installed/authenticated coding tool required; no bundled model quota.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/cli
Locator: Installation; First connection; Common issue operations; Command reference
Finding: CLI manages workspaces, issues, agents, automations and runtimes. OAuth setup/PAT login, shared issue IDs, status/assignment/comments, run inspect/stop/retry documented; installed --help is authoritative for flags.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/assigning-issues
Locator: Assign and start a run; Assign without starting; Assignment permissions; Use the CLI
Finding: Re-read 2026-10-09: Assignment starts a run unless the issue is in backlog or "Don't start yet" is chosen. Access options: Only me, Specific people, Entire workspace. "a run is bound to its runtime and never moves to another machine." CLI assign commands including --no-start.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/inbox
Locator: What lands in the inbox; Tune notification types
Finding: Inbox collects assignment, comment, mention and agent-failure attention events; full history remains on the issue.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/issues
Locator: Searches 2026-10-08: Android, App Store, SSO/SAML, dependency/blocked-by, budget/spend limit, cloud runtime
Finding: Newest issue list reread November 7. Requests are not vendor announcements; prior October counts are not refreshed.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/releases/tag/v0.4.37
Locator: Release body: commit 9c8c7f0 "feat(mobile): enable native iPad support (#7793)"
Finding: Re-read 2026-10-09 via the GitHub API: the v0.4.37 body lists commit 9c8c7f0 "feat(mobile): enable native iPad support (#7793)". The live changelog entry ("Install Multica natively on iPad, in any screen orientation") could not be re-fetched (HTTP 403); it rests on the reviewer read.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/b7f8efcaae6c5f3273c0038252b5d10c81b41068/packages/views/issues/hooks/use-sticky-composer.ts
Locator: useStickyComposer; apps/web browser application
Finding: The shared issue view explicitly adapts its comment composer to phone-sized browser viewports via useIsMobile. The application is exposed through the web app as well as desktop clients.
Retrieved: 2026-11-13

## multica:ACC-02

Finding: The vendor website, the providers table and the install table each list 29 supported coding tools, including ZeroClaw, Muse Code, Devin CLI and ZCode. Multica launches and drives each tool CLI through a local daemon rather than embedding a harness. Custom runtime profiles must fit an existing protocol family. Multica v0.8.2 detects and invokes Antigravity through the installed agy binary, with optional binary-path and model overrides.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop; Pinned Multica source
Version: Live docs v0.8.1 (docs source at tag v0.8.1, commit b1ea927); protocol source at commit 8db6cfe Antigravity source pass: v0.8.2 (b7f8efcaae6c5f3273c0038252b5d10c81b41068).

Documentation and source based; no run with two different tools was observed.
Supported means Multica can invoke the CLI; the tool, its account and model access come from the user.
Install docs set minimum versions for eleven tools (Antigravity 1.1.10, Claude Code 2.0.0, Codex 0.100.0, GitHub Copilot CLI 1.0.0, Grok 0.2.89, Qwen Code 0.20.0, MiniMax Code 0.1.2, OpenCode 1.1.54, Dim 0.3.10, Devin CLI 3000.3.22, ZeroClaw 0.8.0); Muse Code needs 1.4.10 or later. A tool below its minimum is not registered as a runtime.
The user supplies the Antigravity CLI and account; the existing documented minimum CLI version remains applicable.

Source: https://multica.ai/
Locator: Hero; Works with 20+ AI coding tools; Runtimes; Source available; FAQ
Finding: Re-read 2026-10-09: Names 26 coding tools including ZeroClaw. "Start free trial" button, no price list, "no artificial caps when you self-host", source-available wording. Dashboard mockup shows a "Cloud (Anthropic)" runtime, which the download page describes as not live.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/providers
Locator: Supported tools table; Model sources; Session resumption; MCP configuration; Skill injection
Finding: Re-read 2026-10-09: Supported tools table (25 CLIs). Model lists come from each runtime. QwenPaw and MiniMax Code do not accept a model chosen in Multica. No full provider list. Session resumption continues for tools marked as supporting it; MiniMax Code starts a fresh session. MCP is not read for Antigravity, GitHub Copilot CLI, DevEco Code or Pi.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/install-agent-runtime
Locator: Choose a tool; Sign in on your machine
Finding: Re-read 2026-10-09: Sign in to each tool on its own machine; Multica never receives login tokens. "Supported" means Multica can invoke the CLI; it does not supply the tool account, subscription or model quota. The table lists 25 CLIs. Minimum versions for eight tools; below minimum the runtime is not registered.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tutorial
Locator: Prepare the site repository; Reviewer; Merge and close out; Form a squad; Automation
Finding: Re-read 2026-10-09: walkthrough. The first agent uses Claude with Sonnet (line 60); the Engineer agent, created through Multica helper, uses Opus (line 85); Reviewer uses Codex with GPT-5.6 Sol in a fresh context and never edits code (line 208). The Engineer opens the PR using the GitHub CLI on the user's computer (line 95). Merging is requested in a comment. Merge-driven status is configurable: the GitHub integration docs say that after PRs merge the issue moves to Done by default or is left alone (github-integration, "After PRs merge, move the issue to"). A PR that references an issue identifier links automatically (line 107).
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/CLI_AND_DAEMON.md
Locator: Supported Agents table lines 217-250; How It Works lines 251-258; per-tool launch notes lines 355-396
Finding: Re-read 2026-10-09: Pinned CLI_AND_DAEMON.md (8db6cfe): runtime table with protocol families including ACP for several tools; MCP servers passed over ACP session/new and resume requests; MiniMax Code 0.1.2 has no session-loading capability; remote MCP transports depend on declared capability.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/pkg/agent/claude.go
Locator: Execute (lines 36-90); buildClaudeArgs lines 1060-1115
Finding: Re-read 2026-10-09: Pinned claude.go (8db6cfe): runs Claude Code with stream-json input and output and keeps stdin open after the first message. Passes --max-turns from the run MaxTurns option (line 1110).
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/views/runtimes/utils.ts
Locator: MODEL_PRICING table lines 203-400; estimateCost
Finding: Re-read 2026-10-09: Pinned runtimes/utils.ts (8db6cfe): static MODEL_PRICING table of provider rates; estimateCost computes cost from token counts and can prefer per-turn prices.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/pkg/agent/codex.go
Locator: Comments at lines 44-58 (app-server subcommand and turn/started reporting)
Finding: Re-read 2026-10-09: Pinned codex.go (8db6cfe): drives the Codex app-server subcommand and turn events, with turn-interrupt timeouts.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/releases/tag/v0.4.33
Locator: Release body: commit 8a5a6adb "feat(agent): add ZeroClaw as a native ACP runtime (MUL-6511) (#7351)"
Finding: Re-read 2026-10-09 via the GitHub API: v0.4.33 body lists "feat(agent): add ZeroClaw as a native ACP runtime (MUL-6511) (#7351)". ZeroClaw is shipped in this release.
Retrieved: 2026-10-09

Source: https://multica.ai/
Locator: Hero; Works with 20+ AI coding tools; Runtimes; Source available; FAQ
Finding: Homepage now names 29 coding tools, source-available internal self-hosting, a free-trial button with no numeric trial terms, and connected-computer execution. Self-hosted Qoder Cloud Agents is a separately configured remote runtime.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/b1ea927def6490c43bedff537b6f4ed9a47fa828/apps/docs/content/docs/providers.mdx
Locator: Supported tools table lines 22-52; Model sources line 72; MCP configuration lines 88-92
Finding: Docs source behind multica.ai/docs/providers at v0.8.1 lists 29 tools, adding Muse Code, Devin CLI, ZCode and ZeroClaw to the table. QwenPaw, MiniMax Code and ZeroClaw take their model from their own configuration. Antigravity, GitHub Copilot CLI, DevEco Code, Muse Code and ZeroClaw do not read the agent MCP configuration; managed MCP for Pi needs Pi 0.99.0 or later. Dim and ZeroClaw have no skill injection path.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/b1ea927def6490c43bedff537b6f4ed9a47fa828/apps/docs/content/docs/install-agent-runtime.mdx
Locator: Choose a tool table lines 20-52; Minimum version table lines 67-79; Muse Code accordion
Finding: Docs source behind multica.ai/docs/install-agent-runtime at v0.8.1 lists 29 detected commands and minimum versions for 11 tools (Antigravity, Claude Code, Codex, Copilot, Grok, Qwen Code, MiniMax Code, OpenCode, Dim, Devin CLI, ZeroClaw); Muse Code needs 1.4.10 or later. Tools below minimum are not registered.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/b7f8efcaae6c5f3273c0038252b5d10c81b41068/server/internal/daemon/agents_probe.go#L260-L272
Locator: server/internal/daemon/agents_probe.go lines 260-272, v0.8.2, commit b7f8efcaae6c5f3273c0038252b5d10c81b41068
Finding: Multica v0.8.2 detects and invokes Antigravity through the installed agy binary, with optional binary-path and model overrides.
Retrieved: 2026-11-10

Source: https://github.com/multica-ai/multica/blob/b7f8efcaae6c5f3273c0038252b5d10c81b41068/CLI_AND_DAEMON.md#L268-L280
Locator: CLI_AND_DAEMON.md lines 268-280, v0.8.2, commit b7f8efcaae6c5f3273c0038252b5d10c81b41068
Finding: Multica v0.8.2 detects and invokes Antigravity through the installed agy binary, with optional binary-path and model overrides.
Retrieved: 2026-11-10

## multica:ACC-03

Finding: Multica does not provide model access in the sources read: each coding tool uses the account, subscription or API key configured for that tool on the user's machine. Multica Cloud requires a Multica account (email code or Google); a self-hosted server has its own email-code sign-in and needs no Multica Cloud account.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Multica Cloud/web/desktop; Pinned Multica source
Version: Live docs reread 2026-11-07 through v0.8.1; source at v0.6.1 commit 8db6cfe

Per-tool subscription support, local models and account switching were not established.
Existing subscription support cannot be inferred from API-key support.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-11-07

Source: https://multica.ai/download
Locator: Choose your platform; Prefer the CLI?; Cloud runtime (waitlist); Current version
Finding: Downloads show macOS Apple Silicon/Intel, Windows x64/ARM64 and Linux x64/ARM64 installers; current v0.8.1. Vendor-hosted runtime remains not live, waitlist-only. It is separate from connecting Qoder Cloud Agents.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/providers
Locator: Supported tools table; Model sources; Session resumption; MCP configuration; Skill injection
Finding: Current provider table documents local-tool account/model/MCP/skill differences; table demo says 28 while it also includes ZCode and homepage says 29. No universal account compatibility inferred from count.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/cloud-quickstart
Locator: Steps 1-5; Common problems
Finding: Web/Desktop sign-in uses email code or Google. Desktop connects a computer automatically; other computers install CLI and run setup. At least one installed/authenticated coding tool required; no bundled model quota.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/agents-create
Locator: Starting points; Runtime, model, and thinking level; Configuration after creation; Environment variables and credentials
Finding: Current agent config docs: custom_env stays plaintext in server database; owner/admin unlock/read/change is audited and list/detail APIs expose only an opaque count. Agent Access can only be changed by its owner. Runtime/model/thinking/MCP settings have separate scope.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/self-host-quickstart
Locator: Multica service and computer; make selfhost; Back up Postgres; Kubernetes
Finding: Re-read 2026-10-09: Self-hosting runs Web, API and PostgreSQL through Docker Compose; Kubernetes via Helm is referenced. No email service by default; codes are printed to backend logs until Resend or SMTP is set. Migrations are forward-only. "Self-hosting replaces only the Multica Cloud part."
Retrieved: 2026-11-07

Source: https://multica.ai/docs/install-agent-runtime
Locator: Choose a tool; Sign in on your machine
Finding: Re-read 2026-10-09: Sign in to each tool on its own machine; Multica never receives login tokens. "Supported" means Multica can invoke the CLI; it does not supply the tool account, subscription or model quota. The table lists 25 CLIs. Minimum versions for eight tools; below minimum the runtime is not registered.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/core/types/billing.ts
Locator: Header comment lines 1-12; BillingBatch; BillingTopup
Finding: Re-read 2026-10-09: Pinned billing types (8db6cfe): credit wallet with 1 USD = 1000 credits, credit batches and top-up states. Header states a separate multica-cloud billing service. No public description of what credits purchase.
Retrieved: 2026-11-07

## multica:ACC-04

Finding: Documented onboarding is: sign in, connect a computer with Desktop or multica setup, create an agent, assign the first issue. Local non-git directories, local git repositories and any reachable Git URL are supported as project resources; first-task time was not measured.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

No fresh-install test was run; steps are from documentation.
Cloud runtime onboarding is not available (waitlist).

Source: https://multica.ai/download
Locator: Choose your platform; Prefer the CLI?; Cloud runtime (waitlist); Current version
Finding: Re-read 2026-10-09: Current version v0.6.1. Installers: macOS Apple Silicon and Intel (.dmg/.zip), Windows x64 and ARM64 (.exe), Linux x64 and ARM64 (.AppImage, .deb, .rpm). CLI install over SSH. Cloud runtime shown as waitlist, "Not live yet". No iOS, Android or minimum OS statement.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/desktop-app
Locator: Installation; Built-in daemon; Updating Desktop; Connecting to a self-hosted instance
Finding: Re-read 2026-10-09: Installers: macOS (.dmg), Windows (.exe), Linux (.AppImage, .deb, .rpm). Desktop starts a built-in daemon after sign-in. Auto-update on by default. Self-hosted connection through desktop.json. No minimum OS version stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cloud-quickstart
Locator: Steps 1-5; Common problems
Finding: Re-read 2026-10-09: Sign in by email code or Google on web or Desktop. Connect a computer: Desktop registers it automatically; web uses an install script then multica setup. Create an agent, create an issue and submit. Common problems: offline runtime, "Keep Desktop open".
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/project-resources
Locator: Add a GitHub repository; Add a local directory; How runs share the directory
Finding: Re-read 2026-10-09: GitHub repositories are checked out per run through worktrees in a runtime-managed workspace. Local git directories run in place (serialized, directory lock) or in Parallel worktree mode on agent/<agent>/<issue> branches. Non-git directories run serially. Dependency caches, snapshots and port allocation are not mentioned.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/assigning-issues
Locator: Assign and start a run; Assign without starting; Assignment permissions; Use the CLI
Finding: Re-read 2026-10-09: Assignment starts a run unless the issue is in backlog or "Don't start yet" is chosen. Access options: Only me, Specific people, Entire workspace. "a run is bound to its runtime and never moves to another machine." CLI assign commands including --no-start.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/mentioning-agents
Locator: Trigger preview; Automatic routing of replies; Coalescing consecutive comments; Access restrictions
Finding: Re-read 2026-10-09: Mentions and replies start runs; a trigger preview is shown before sending. Simultaneous mentions of different agents are not merged; consecutive comments to one agent merge into the pending run. Runs wait when the runtime is offline. Access settings apply to mentions. No presence indicators.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/self-host-quickstart
Locator: Multica service and computer; make selfhost; Back up Postgres; Kubernetes
Finding: Re-read 2026-10-09: Self-hosting runs Web, API and PostgreSQL through Docker Compose; Kubernetes via Helm is referenced. No email service by default; codes are printed to backend logs until Resend or SMTP is set. Migrations are forward-only. "Self-hosting replaces only the Multica Cloud part."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/install-agent-runtime
Locator: Choose a tool; Sign in on your machine
Finding: Re-read 2026-10-09: Sign in to each tool on its own machine; Multica never receives login tokens. "Supported" means Multica can invoke the CLI; it does not supply the tool account, subscription or model quota. The table lists 25 CLIs. Minimum versions for eight tools; below minimum the runtime is not registered.
Retrieved: 2026-10-09

## multica:ACC-05

Finding: The CLI exposes broad issue, agent, project, autopilot and runtime operations over the shared server API, authenticated by personal access tokens; a REST API is used in documentation but no general API reference or MCP server was found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop; Pinned Multica source
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Public SDK and general REST API stability unknown; MCP server exposure not found.
Deployed parity between source and hosted service was not tested.

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/cmd/multica/cmd_issue.go
Locator: issueCmd subcommands lines 222-456 (list/get/pull-requests/children/create/update/assign/status/comment/runs/usage/rerun/cancel-task/search); comment update --expected-revision lines 725 and 2584-2613
Finding: Re-read 2026-10-09: Pinned source cmd_issue.go (8db6cfe): issue subcommands include list, get, pull-requests, children, create, update, assign, status, comment, runs, run-messages, usage, rerun, cancel-task and search; comment update requires --expected-revision; calls go through the shared server API.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cli
Locator: Installation; First connection; Common issue operations; Command reference
Finding: Re-read 2026-10-09: Commands cover issues, projects, labels, agents, skills, autopilots, workspaces, repos, daemon and runtimes. `multica issue cancel-task <task-id> --issue <issue-id>` cancels a run. `multica issue rerun <id>` re-enqueues for the current assignee. Install via curl script, Homebrew or PowerShell. `issue comment update` requires --expected-revision (source, cmd_issue.go).
Retrieved: 2026-10-09

Source: https://multica.ai/docs/autopilots
Locator: Create; Execution mode; Run on a schedule; Run from a webhook; Failures, pausing, and deletion; Permissions
Finding: Re-read 2026-10-09: Autopilots use 5-field cron with IANA time zones, several schedules per autopilot, and run-now. Webhook triggers (JSON up to 256 KiB, idempotency key, GitHub, GitLab and generic event inference, event filters, replay). Create-issue and run-only modes. Run-only failures are not retried and the next schedule still fires. Create-issue waits for an offline runtime; run-only is skipped. Manual pause stops schedules, webhooks and run-now. Automatic pause after at least 50 completed or failed runs in 7 days with a 90% failure rate. The page does not say whether runs can overlap.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/auth-tokens
Locator: Personal access tokens; Using a PAT in API requests; Temporary tokens for agent runs
Finding: Re-read 2026-10-09: Browser sessions last 30 days. Personal access tokens (mul_ prefix, 30 days, 90 days, 1 year or never) work against REST endpoints. multica login creates a 90-day token. Run-scoped mat_ tokens last at most 24 hours and attribute actions to the agent. No dedicated REST API reference.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/featureflags/keys.go
Locator: BillingWorkspaceSubscriptions, ComposioMCPApps, PluginsV1, TriageV1 (lines 8-60)
Finding: Re-read 2026-10-09: Pinned featureflags/keys.go (8db6cfe): billing_workspace_subscriptions defaults to off and gates workspace entitlement, Stripe Checkout and seat reconcile; composio_mcp_apps, plugins_v1 and triage_v1 are other gates.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/pkg/publicapi/v1/openapi.yaml
Locator: servers; /issues/{issue_ref}; /issues/{issue_ref}/comments; /storage
Finding: Re-read 2026-10-09: Pinned source publicapi/v1/openapi.yaml (8db6cfe): OpenAPI v1 covers plugin-token access to context, issues, comments and storage. It is a plugin contract, not a general REST reference.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/plugin-sdk/README.md
Locator: What a surface is; Publishing; What you can do, and what bounds it
Finding: Re-read 2026-10-09: Pinned source plugin-sdk README (8db6cfe): plugin surfaces run in a sandboxed iframe with allow-scripts only; they read issues, post comments authored by the user without @-mention dispatch, and use scoped storage. Catalog gated by plugins_v1.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/issues
Locator: Searches 2026-10-08: Android, App Store, SSO/SAML, dependency/blocked-by, budget/spend limit, cloud runtime
Finding: Re-read 2026-10-09: GitHub searches on 2026-10-09 return 16 open Android results, 12 App Store results and 13 SSO results. Open community requests: #3864 and #3412 (Android), #6996 (mobile push), #7794 and #1218 (blocker dependencies), #6950 (spending limit), #5423 (Azure AAD), #4151 (open-source the hosted runtime). None is a vendor announcement. The repository reports 1,839 open issues and pull requests combined.
Retrieved: 2026-10-09

## multica:ACC-06

Finding: Workspace owners/admins keep an MCP server library that is assigned per agent and passed to most supported tools; skills import from GitHub, ClawHub, Skills.sh, local archives or a connected computer. Setup is product-managed for MCP and skills, while some tools keep their own MCP configuration.

Verification: sourced
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: Live docs v0.8.1 (docs source at tag v0.8.1, commit b1ea927); feature flags at commit 8db6cfe

No external-tool call and access boundary was verified.
Custom tools beyond MCP and skills were not found.

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/providers
Locator: Supported tools table; Model sources; Session resumption; MCP configuration; Skill injection
Finding: Re-read 2026-10-09: Supported tools table (25 CLIs). Model lists come from each runtime. QwenPaw and MiniMax Code do not accept a model chosen in Multica. No full provider list. Session resumption continues for tools marked as supporting it; MiniMax Code starts a fresh session. MCP is not read for Antigravity, GitHub Copilot CLI, DevEco Code or Pi.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cli
Locator: Installation; First connection; Common issue operations; Command reference
Finding: Re-read 2026-10-09: Commands cover issues, projects, labels, agents, skills, autopilots, workspaces, repos, daemon and runtimes. `multica issue cancel-task <task-id> --issue <issue-id>` cancels a run. `multica issue rerun <id>` re-enqueues for the current assignee. Install via curl script, Homebrew or PowerShell. `issue comment update` requires --expected-revision (source, cmd_issue.go).
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents-create
Locator: Starting points; Runtime, model, and thinking level; Configuration after creation; Environment variables and credentials
Finding: Re-read 2026-10-09: Agents start blank or with Build with AI (needs an online runtime). Runtime, model and thinking level. custom_env values are stored in plaintext in the Multica server database; only workspace owners and admins can view or change them. Critical variables such as PATH, HOME and MULTICA_* cannot be overridden.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/skills
Locator: Creating or importing a skill; Updating an imported skill; Workspace skills and repository skills; Edit permissions
Finding: Re-read 2026-10-09: Skills are SKILL.md bundles, created manually or imported from a local folder or archive, GitHub, ClawHub, Skills.sh, or copied from a connected runtime. Skills attach per agent; changes apply from the next run. No review or sandbox of imported content stated.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/featureflags/keys.go
Locator: BillingWorkspaceSubscriptions, ComposioMCPApps, PluginsV1, TriageV1 (lines 8-60)
Finding: Re-read 2026-10-09: Pinned featureflags/keys.go (8db6cfe): billing_workspace_subscriptions defaults to off and gates workspace entitlement, Stripe Checkout and seat reconcile; composio_mcp_apps, plugins_v1 and triage_v1 are other gates.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/providers
Locator: Supported tools; Model sources; MCP configuration
Finding: Multica invokes installed signed-in coding tools, including Claude Code, Codex, Cursor Agent, Grok, OpenCode and Pi. MCP custom tools follow each runtime’s documented configuration support.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/b1ea927def6490c43bedff537b6f4ed9a47fa828/apps/docs/content/docs/providers.mdx
Locator: Supported tools table lines 22-52; Model sources line 72; MCP configuration lines 88-92
Finding: Docs source behind multica.ai/docs/providers at v0.8.1 lists 29 tools, adding Muse Code, Devin CLI, ZCode and ZeroClaw to the table. QwenPaw, MiniMax Code and ZeroClaw take their model from their own configuration. Antigravity, GitHub Copilot CLI, DevEco Code, Muse Code and ZeroClaw do not read the agent MCP configuration; managed MCP for Pi needs Pi 0.99.0 or later. Dim and ZeroClaw have no skill injection path.
Retrieved: 2026-11-07

## multica:ACC-07

Finding: Desktop uses Electron; mobile docs specify React Native.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2027-01-07
Plan: Not specified
Surface: Pinned Multica source; Self-built iOS client
Version: commit 8db6cfe

Installed binary parity with the pinned source is unverified.
Hybrid structure (shared web UI inside Electron) was not inspected.

Source: https://multica.ai/docs/mobile-app
Locator: Intro; Prerequisites; 7-day limit of free signing
Finding: Re-read 2026-10-09: The iOS client is "not on the App Store yet". Build a Release from the public repository on a Mac with Xcode and install over USB. Free Apple ID signing "only work for 7 days". The self-installed build does not auto-update. "There is no Android client yet." No iPad statement. No feature list.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/apps/desktop/package.json
Locator: dependencies: electron ^39.2.6, electron-builder ^26.0.12, electron-vite ^5.0.0 (lines 72-74)
Finding: Re-read 2026-10-09: Pinned source apps/desktop/package.json (8db6cfe): Electron ^39.2.6, electron-builder ^26.0.12, electron-vite ^5.0.0. The Desktop app is an Electron build.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/apps/mobile/package.json
Locator: dependencies expo ~55.0.23 (line 46), react-native 0.83.6 (line 68)
Finding: Re-read 2026-10-09: Pinned source apps/mobile/package.json (8db6cfe): Expo ~55.0.23 and react-native 0.83.6; the mobile client is an Expo and React Native app.
Retrieved: 2026-10-09

## multica:PLAN-01

Finding: Multica issues are the native record: issues are created in the app, CLI, chat bots or by prompting an agent. GitHub is linked only for pull requests, read-only; no import from Linear, GitHub Issues or Jira was found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

External issue import, edit sync and round trips were not found; an open GitHub issues request exists (#8822).
Documentation based; no intake scenario was run.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cloud-quickstart
Locator: Steps 1-5; Common problems
Finding: Re-read 2026-10-09: Sign in by email code or Google on web or Desktop. Connect a computer: Desktop registers it automatically; web uses an install script then multica setup. Create an agent, create an issue and submit. Common problems: offline runtime, "Keep Desktop open".
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cli
Locator: Installation; First connection; Common issue operations; Command reference
Finding: Re-read 2026-10-09: Commands cover issues, projects, labels, agents, skills, autopilots, workspaces, repos, daemon and runtimes. `multica issue cancel-task <task-id> --issue <issue-id>` cancels a run. `multica issue rerun <id>` re-enqueues for the current assignee. Install via curl script, Homebrew or PowerShell. `issue comment update` requires --expected-revision (source, cmd_issue.go).
Retrieved: 2026-10-09

Source: https://multica.ai/docs/autopilots
Locator: Create; Execution mode; Run on a schedule; Run from a webhook; Failures, pausing, and deletion; Permissions
Finding: Re-read 2026-10-09: Autopilots use 5-field cron with IANA time zones, several schedules per autopilot, and run-now. Webhook triggers (JSON up to 256 KiB, idempotency key, GitHub, GitLab and generic event inference, event filters, replay). Create-issue and run-only modes. Run-only failures are not retried and the next schedule still fires. Create-issue waits for an offline runtime; run-only is skipped. Manual pause stops schedules, webhooks and run-now. Automatic pause after at least 50 completed or failed runs in 7 days with a 90% failure rate. The page does not say whether runs can overlap.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/chat
Locator: Intro; Multi-turn context; Watch and stop a run
Finding: Re-read 2026-10-09: Private one-to-one chat with an agent outside issues. Attachments supported ("attach images or files"). Session continues where possible. Stopping a run restores unprocessed input to the editor. Offline runtime makes messages wait. No code search, indexing or retrieval statement.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/channels
Locator: Choose a platform; Account binding; Self-hosting
Finding: Re-read 2026-10-09: Feishu/Lark, Slack, DingTalk, WeCom and Telegram. "New connections are currently open only for mainland-China Feishu." Account binding on first message; membership re-checked on every message.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/workspaces
Locator: What a workspace contains; Delete a workspace
Finding: Re-read 2026-10-09: A workspace is the isolation boundary for members, issues, projects, agents, skills and run history. Owners and admins manage settings; members view them. Deletion is permanent. No export or portability statement. No SSO mention.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/issues
Locator: Searches 2026-10-08: Android, App Store, SSO/SAML, dependency/blocked-by, budget/spend limit, cloud runtime
Finding: Re-read 2026-10-09: GitHub searches on 2026-10-09 return 16 open Android results, 12 App Store results and 13 SSO results. Open community requests: #3864 and #3412 (Android), #6996 (mobile push), #7794 and #1218 (blocker dependencies), #6950 (spending limit), #5423 (Azure AAD), #4151 (open-source the hosted runtime). None is a vendor announcement. The repository reports 1,839 open issues and pull requests combined.
Retrieved: 2026-10-09

## multica:PLAN-02

Finding: Issues hold the task brief and requirements (description text, comments, attachments); no separate structured plan object or acceptance-criteria field was found in docs or the CLI.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

No test of amendment while running was executed.
Editors for plans and acceptance criteria not found as a separate feature.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/cmd/multica/cmd_issue.go
Locator: issueCmd subcommands lines 222-456 (list/get/pull-requests/children/create/update/assign/status/comment/runs/usage/rerun/cancel-task/search); comment update --expected-revision lines 725 and 2584-2613
Finding: Re-read 2026-10-09: Pinned source cmd_issue.go (8db6cfe): issue subcommands include list, get, pull-requests, children, create, update, assign, status, comment, runs, run-messages, usage, rerun, cancel-task and search; comment update requires --expected-revision; calls go through the shared server API.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/handler/issue.go
Locator: IssueResponse DTO lines 40-90 (status, parent_issue_id, stage, revision, properties); status key resolution via workspace catalog lines 185-200
Finding: Re-read 2026-10-09: Pinned source internal/handler/issue.go (8db6cfe): the issue response carries parent_issue_id, stage, revision, custom properties and status; status keys resolve against the workspace status catalog.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/mentioning-agents
Locator: Trigger preview; Automatic routing of replies; Coalescing consecutive comments; Access restrictions
Finding: Re-read 2026-10-09: Mentions and replies start runs; a trigger preview is shown before sending. Simultaneous mentions of different agents are not merged; consecutive comments to one agent merge into the pending run. Runs wait when the runtime is offline. Access settings apply to mentions. No presence indicators.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/handler/activity.go
Locator: TimelineEntry lines 14-50
Finding: Re-read 2026-10-09: Pinned source internal/handler/activity.go (8db6cfe): the issue timeline interleaves activity and comments with durable actor_type and actor_id attribution; deleted thread roots keep a tombstone.
Retrieved: 2026-10-09

## multica:PLAN-03

Finding: Issues have parent-child relationships.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Blocker dependencies are not documented and no CLI command uses them; open requests #7794 and #1218 ask for blocked_by with dispatch admission.
Dependency enforcement, readiness and identity retention while growing a task were not tested.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/handler/issue.go
Locator: IssueResponse DTO lines 40-90 (status, parent_issue_id, stage, revision, properties); status key resolution via workspace catalog lines 185-200
Finding: Re-read 2026-10-09: Pinned source internal/handler/issue.go (8db6cfe): the issue response carries parent_issue_id, stage, revision, custom properties and status; status keys resolve against the workspace status catalog.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/projects
Locator: Parts of a project; Project description and execution context; Project lead
Finding: Re-read 2026-10-09: Projects group issues (one project per issue) with status, priority, lead, dates, description and resources. The description enters agent execution context. Lead is not a permission and does not auto-assign or run agents. No code search or indexing statement.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/issues
Locator: Searches 2026-10-08: Android, App Store, SSO/SAML, dependency/blocked-by, budget/spend limit, cloud runtime
Finding: Re-read 2026-10-09: GitHub searches on 2026-10-09 return 16 open Android results, 12 App Store results and 13 SSO results. Open community requests: #3864 and #3412 (Android), #6996 (mobile push), #7794 and #1218 (blocker dependencies), #6950 (spending limit), #5423 (Azure AAD), #4151 (open-source the hosted runtime). None is a vendor announcement. The repository reports 1,839 open issues and pull requests combined.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/workflows
Locator: Agent steps; Human steps; Handoffs
Finding: Workflow columns can hand issues to an agent or person, define the next state after completion and a send-back path. Blocked is reserved for outside obstacles.
Retrieved: 2026-11-07

## multica:PLAN-04

Finding: Agent instructions, workspace context, project descriptions and reusable skills provide shared context; sessions resume where the tool supports it. No managed cross-harness memory or stale-context controls were found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Managed memory across sessions or harnesses was not found; tool-native memory is the tool's.
No new-session or colleague takeover test was run.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/providers
Locator: Supported tools table; Model sources; Session resumption; MCP configuration; Skill injection
Finding: Re-read 2026-10-09: Supported tools table (25 CLIs). Model lists come from each runtime. QwenPaw and MiniMax Code do not accept a model chosen in Multica. No full provider list. Session resumption continues for tools marked as supporting it; MiniMax Code starts a fresh session. MCP is not read for Antigravity, GitHub Copilot CLI, DevEco Code or Pi.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents
Locator: Agent configuration; Permissions and Access; Online, offline, and workload status
Finding: Re-read 2026-10-09: Agent settings: name, instructions, skills, runtime, model and thinking level, Access (Only me default, Specific people, Entire workspace), concurrency, env, MCP. Availability comes from the runtime and workload from runs. Archiving cancels unfinished runs.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/skills
Locator: Creating or importing a skill; Updating an imported skill; Workspace skills and repository skills; Edit permissions
Finding: Re-read 2026-10-09: Skills are SKILL.md bundles, created manually or imported from a local folder or archive, GitHub, ClawHub, Skills.sh, or copied from a connected runtime. Skills attach per agent; changes apply from the next run. No review or sandbox of imported content stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/projects
Locator: Parts of a project; Project description and execution context; Project lead
Finding: Re-read 2026-10-09: Projects group issues (one project per issue) with status, priority, lead, dates, description and resources. The description enters agent execution context. Lead is not a permission and does not auto-assign or run agents. No code search or indexing statement.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/chat
Locator: Intro; Multi-turn context; Watch and stop a run
Finding: Re-read 2026-10-09: Private one-to-one chat with an agent outside issues. Attachments supported ("attach images or files"). Session continues where possible. Stopping a run restores unprocessed input to the editor. Offline runtime makes messages wait. No code search, indexing or retrieval statement.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/workspaces
Locator: What a workspace contains; Delete a workspace
Finding: Re-read 2026-10-09: A workspace is the isolation boundary for members, issues, projects, agents, skills and run history. Owners and admins manage settings; members view them. Deletion is permanent. No export or portability statement. No SSO mention.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Codex native memory; Sessions, memory and temporary directories
Finding: Codex native memory is off by default per run and can be enabled with MULTICA_CODEX_MEMORY. Hermes memory is per agent with a 90-day unused retention policy.
Retrieved: 2026-11-07

## multica:PLAN-05

Finding: Multica passes attachments and chat media to agents but no product-level code search, indexing, architecture retrieval or conflict detection was found; code understanding comes from the underlying coding tool.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Everything except multimodal inputs was not established; no architecture-conflict scenario was run.

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/chat
Locator: Intro; Multi-turn context; Watch and stop a run
Finding: Re-read 2026-10-09: Private one-to-one chat with an agent outside issues. Attachments supported ("attach images or files"). Session continues where possible. Stopping a run restores unprocessed input to the editor. Offline runtime makes messages wait. No code search, indexing or retrieval statement.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/channels
Locator: Choose a platform; Account binding; Self-hosting
Finding: Re-read 2026-10-09: Feishu/Lark, Slack, DingTalk, WeCom and Telegram. "New connections are currently open only for mainland-China Feishu." Account binding on first message; membership re-checked on every message.
Retrieved: 2026-10-09

## multica:PLAN-06

Finding: Native issues include creation, editing, comments, priority, labels, assignment, search, filters, projects, five views, backlog parking and explicit dispatch with separate retained runs.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Custom unstarted statuses do not inherit backlog suppression.
Issue archive was not found (deletion is permanent; cancelled keeps the record); the full save-refine-dispatch scenario with a second person was not run.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/cmd/multica/cmd_issue.go
Locator: issueCmd subcommands lines 222-456 (list/get/pull-requests/children/create/update/assign/status/comment/runs/usage/rerun/cancel-task/search); comment update --expected-revision lines 725 and 2584-2613
Finding: Re-read 2026-10-09: Pinned source cmd_issue.go (8db6cfe): issue subcommands include list, get, pull-requests, children, create, update, assign, status, comment, runs, run-messages, usage, rerun, cancel-task and search; comment update requires --expected-revision; calls go through the shared server API.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cli
Locator: Installation; First connection; Common issue operations; Command reference
Finding: Re-read 2026-10-09: Commands cover issues, projects, labels, agents, skills, autopilots, workspaces, repos, daemon and runtimes. `multica issue cancel-task <task-id> --issue <issue-id>` cancels a run. `multica issue rerun <id>` re-enqueues for the current assignee. Install via curl script, Homebrew or PowerShell. `issue comment update` requires --expected-revision (source, cmd_issue.go).
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/projects
Locator: Parts of a project; Project description and execution context; Project lead
Finding: Re-read 2026-10-09: Projects group issues (one project per issue) with status, priority, lead, dates, description and resources. The description enters agent execution context. Lead is not a permission and does not auto-assign or run agents. No code search or indexing statement.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/assigning-issues
Locator: Assign and start a run; Assign without starting; Assignment permissions; Use the CLI
Finding: Re-read 2026-10-09: Assignment starts a run unless the issue is in backlog or "Don't start yet" is chosen. Access options: Only me, Specific people, Entire workspace. "a run is bound to its runtime and never moves to another machine." CLI assign commands including --no-start.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/comments
Locator: Post comments; Annotate; Resolve a discussion; @-mentions; Edit and delete
Finding: Re-read 2026-10-09: Threaded comments with attachments, annotations, resolve or conclusion, mentions and reactions. Authors edit or delete their own comments; owners and admins edit or delete anyone's; agent-posted comments can only be deleted. Deletion is permanent. Drafts expire after 30 days. Retention and export not stated.
Retrieved: 2026-10-09

## multica:PLAN-07

Finding: Agents use CLI against shared issue records.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Pinned Multica source; Multica Cloud/web/desktop
Version: commit 8db6cfe; v0.6.1

Claiming work and dependency creation by agents were not found.
Issue-wide locking and permission end-to-end tests remain open.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/cmd/multica/cmd_issue.go
Locator: issueCmd subcommands lines 222-456 (list/get/pull-requests/children/create/update/assign/status/comment/runs/usage/rerun/cancel-task/search); comment update --expected-revision lines 725 and 2584-2613
Finding: Re-read 2026-10-09: Pinned source cmd_issue.go (8db6cfe): issue subcommands include list, get, pull-requests, children, create, update, assign, status, comment, runs, run-messages, usage, rerun, cancel-task and search; comment update requires --expected-revision; calls go through the shared server API.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/handler/issue.go
Locator: IssueResponse DTO lines 40-90 (status, parent_issue_id, stage, revision, properties); status key resolution via workspace catalog lines 185-200
Finding: Re-read 2026-10-09: Pinned source internal/handler/issue.go (8db6cfe): the issue response carries parent_issue_id, stage, revision, custom properties and status; status keys resolve against the workspace status catalog.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cli
Locator: Installation; First connection; Common issue operations; Command reference
Finding: Re-read 2026-10-09: Commands cover issues, projects, labels, agents, skills, autopilots, workspaces, repos, daemon and runtimes. `multica issue cancel-task <task-id> --issue <issue-id>` cancels a run. `multica issue rerun <id>` re-enqueues for the current assignee. Install via curl script, Homebrew or PowerShell. `issue comment update` requires --expected-revision (source, cmd_issue.go).
Retrieved: 2026-10-09

Source: https://multica.ai/docs/assigning-issues
Locator: Assign and start a run; Assign without starting; Assignment permissions; Use the CLI
Finding: Re-read 2026-10-09: Assignment starts a run unless the issue is in backlog or "Don't start yet" is chosen. Access options: Only me, Specific people, Entire workspace. "a run is bound to its runtime and never moves to another machine." CLI assign commands including --no-start.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/auth-tokens
Locator: Personal access tokens; Using a PAT in API requests; Temporary tokens for agent runs
Finding: Re-read 2026-10-09: Browser sessions last 30 days. Personal access tokens (mul_ prefix, 30 days, 90 days, 1 year or never) work against REST endpoints. multica login creates a 90-day token. Run-scoped mat_ tokens last at most 24 hours and attribute actions to the agent. No dedicated REST API reference.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/handler/activity.go
Locator: TimelineEntry lines 14-50
Finding: Re-read 2026-10-09: Pinned source internal/handler/activity.go (8db6cfe): the issue timeline interleaves activity and comments with durable actor_type and actor_id attribution; deleted thread roots keep a tombstone.
Retrieved: 2026-10-09

## multica:PLAN-08

Finding: Issue descriptions can hold persistent requirements editable by people and agents. A dedicated living specification, revision approval and drift checking remain unestablished.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-12-07
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: Multica docs and v0.8.1 documentation inspected 2026-11-07

Requirements live in issue descriptions; specification revision history, code links and drift detection are not established.

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/cmd/multica/cmd_issue.go
Locator: issueCmd subcommands lines 222-456 (list/get/pull-requests/children/create/update/assign/status/comment/runs/usage/rerun/cancel-task/search); comment update --expected-revision lines 725 and 2584-2613
Finding: Re-read 2026-10-09: Pinned source cmd_issue.go (8db6cfe): issue subcommands include list, get, pull-requests, children, create, update, assign, status, comment, runs, run-messages, usage, rerun, cancel-task and search; comment update requires --expected-revision; calls go through the shared server API.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Create an issue
Finding: An issue description contains goals, requirements and acceptance criteria; discussions, state changes and execution history remain with the issue.
Retrieved: 2026-11-07

## multica:COORD-01

Finding: Multiple runs execute concurrently on a daemon (default 20 per daemon and 6 per agent) and work state is visible in the agent list, issue execution log, header chip and runtime page.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Observed concurrency was not tested; squads do not raise concurrency by themselves.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/squads
Locator: Squad composition; Execution flow after assignment; Leader re-trigger rules
Finding: Re-read 2026-10-09: Assignment to a squad wakes only the leader, which @-mentions members, records an evaluation and stops. The leader is woken again when a delegated member replies, a non-member comments or a member posts progress without a mention. Members can be agents or humans. Provider-native subagents are not described.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents
Locator: Agent configuration; Permissions and Access; Online, offline, and workload status
Finding: Re-read 2026-10-09: Agent settings: name, instructions, skills, runtime, model and thinking level, Access (Only me default, Specific people, Entire workspace), concurrency, env, MCP. Availability comes from the runtime and workload from runs. Archiving cancels unfinished runs.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents-create
Locator: Starting points; Runtime, model, and thinking level; Configuration after creation; Environment variables and credentials
Finding: Re-read 2026-10-09: Agents start blank or with Build with AI (needs an online runtime). Runtime, model and thinking level. custom_env values are stored in plaintext in the Multica server database; only workspace owners and admins can view or change them. Critical variables such as PATH, HOME and MULTICA_* cannot be overridden.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/environment-variables
Locator: Daemon configuration; Observability and analytics; Server-side LLM
Finding: Re-read 2026-10-09: Telemetry on by default: one anonymous self-host snapshot per UTC day; DO_NOT_TRACK=1 or true disables it; ANALYTICS_DISABLED=true turns off PostHog. MULTICA_DAEMON_MAX_CONCURRENT_TASKS default 20. MULTICA_AGENT_TIMEOUT default 0 (no limit). MULTICA_AGENT_IDLE_WATCHDOG default 2h; tool watchdog same by default. No iteration limit variable listed. Optional server-side LLM for chat titles sends chat content to a configured endpoint.
Retrieved: 2026-10-09

## multica:COORD-02

Finding: Issues can be assigned to members, agents or squads; runs queue for a runtime and are claimed by it; consecutive triggers coalesce and a local directory lock can defer execution.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Concurrent duplicate-claim behavior not tested.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/squads
Locator: Squad composition; Execution flow after assignment; Leader re-trigger rules
Finding: Re-read 2026-10-09: Assignment to a squad wakes only the leader, which @-mentions members, records an evaluation and stops. The leader is woken again when a delegated member replies, a non-member comments or a member posts progress without a mention. Members can be agents or humans. Provider-native subagents are not described.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents
Locator: Agent configuration; Permissions and Access; Online, offline, and workload status
Finding: Re-read 2026-10-09: Agent settings: name, instructions, skills, runtime, model and thinking level, Access (Only me default, Specific people, Entire workspace), concurrency, env, MCP. Availability comes from the runtime and workload from runs. Archiving cancels unfinished runs.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/project-resources
Locator: Add a GitHub repository; Add a local directory; How runs share the directory
Finding: Re-read 2026-10-09: GitHub repositories are checked out per run through worktrees in a runtime-managed workspace. Local git directories run in place (serialized, directory lock) or in Parallel worktree mode on agent/<agent>/<issue> branches. Non-git directories run serially. Dependency caches, snapshots and port allocation are not mentioned.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/assigning-issues
Locator: Assign and start a run; Assign without starting; Assignment permissions; Use the CLI
Finding: Re-read 2026-10-09: Assignment starts a run unless the issue is in backlog or "Don't start yet" is chosen. Access options: Only me, Specific people, Entire workspace. "a run is bound to its runtime and never moves to another machine." CLI assign commands including --no-start.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/mentioning-agents
Locator: Trigger preview; Automatic routing of replies; Coalescing consecutive comments; Access restrictions
Finding: Re-read 2026-10-09: Mentions and replies start runs; a trigger preview is shown before sending. Simultaneous mentions of different agents are not merged; consecutive comments to one agent merge into the pending run. Runs wait when the runtime is offline. Access settings apply to mentions. No presence indicators.
Retrieved: 2026-10-09

## multica:COORD-03

Finding: Squads provide product-owned delegation: an agent leader receives the issue, @-mentions member agents, records an evaluation, and is re-triggered by replies; members can run on different tools. No provider-native subagent claims were evaluated.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Child identity, live progress and reopen behavior were not observed; invocation specifics beyond mention markdown not tested.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/squads
Locator: Squad composition; Execution flow after assignment; Leader re-trigger rules
Finding: Re-read 2026-10-09: Assignment to a squad wakes only the leader, which @-mentions members, records an evaluation and stops. The leader is woken again when a delegated member replies, a non-member comments or a member posts progress without a mention. Members can be agents or humans. Provider-native subagents are not described.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents
Locator: Agent configuration; Permissions and Access; Online, offline, and workload status
Finding: Re-read 2026-10-09: Agent settings: name, instructions, skills, runtime, model and thinking level, Access (Only me default, Specific people, Entire workspace), concurrency, env, MCP. Availability comes from the runtime and workload from runs. Archiving cancels unfinished runs.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents-create
Locator: Starting points; Runtime, model, and thinking level; Configuration after creation; Environment variables and credentials
Finding: Re-read 2026-10-09: Agents start blank or with Build with AI (needs an online runtime). Runtime, model and thinking level. custom_env values are stored in plaintext in the Multica server database; only workspace owners and admins can view or change them. Critical variables such as PATH, HOME and MULTICA_* cannot be overridden.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/mentioning-agents
Locator: Trigger preview; Automatic routing of replies; Coalescing consecutive comments; Access restrictions
Finding: Re-read 2026-10-09: Mentions and replies start runs; a trigger preview is shown before sending. Simultaneous mentions of different agents are not merged; consecutive comments to one agent merge into the pending run. Runs wait when the runtime is offline. Access settings apply to mentions. No presence indicators.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/auth-tokens
Locator: Personal access tokens; Using a PAT in API requests; Temporary tokens for agent runs
Finding: Re-read 2026-10-09: Browser sessions last 30 days. Personal access tokens (mul_ prefix, 30 days, 90 days, 1 year or never) work against REST endpoints. multica login creates a 90-day token. Run-scoped mat_ tokens last at most 24 hours and attribute actions to the agent. No dedicated REST API reference.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tutorial
Locator: Prepare the site repository; Reviewer; Merge and close out; Form a squad; Automation
Finding: Re-read 2026-10-09: walkthrough. The first agent uses Claude with Sonnet (line 60); the Engineer agent, created through Multica helper, uses Opus (line 85); Reviewer uses Codex with GPT-5.6 Sol in a fresh context and never edits code (line 208). The Engineer opens the PR using the GitHub CLI on the user's computer (line 95). Merging is requested in a comment. Merge-driven status is configurable: the GitHub integration docs say that after PRs merge the issue moves to Done by default or is left alone (github-integration, "After PRs merge, move the issue to"). A PR that references an issue identifier links automatically (line 107).
Retrieved: 2026-10-09

## multica:COORD-04

Finding: Autopilots run on 5-field cron schedules or webhook triggers from the Multica server, and issue wakeup rules can re-wake an agent on a schedule, comment, status change, sub-issue completion or pull request change, with expiry.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

No schedule or wakeup scenario was run.
An autopilot pauses itself after at least 50 completed or failed runs in seven days with a 90% failure rate, and its creator must resume it.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cloud-quickstart
Locator: Steps 1-5; Common problems
Finding: Re-read 2026-10-09: Sign in by email code or Google on web or Desktop. Connect a computer: Desktop registers it automatically; web uses an install script then multica setup. Create an agent, create an issue and submit. Common problems: offline runtime, "Keep Desktop open".
Retrieved: 2026-10-09

Source: https://multica.ai/docs/autopilots
Locator: Create; Execution mode; Run on a schedule; Run from a webhook; Failures, pausing, and deletion; Permissions
Finding: Re-read 2026-10-09: Autopilots use 5-field cron with IANA time zones, several schedules per autopilot, and run-now. Webhook triggers (JSON up to 256 KiB, idempotency key, GitHub, GitLab and generic event inference, event filters, replay). Create-issue and run-only modes. Run-only failures are not retried and the next schedule still fires. Create-issue waits for an offline runtime; run-only is skipped. Manual pause stops schedules, webhooks and run-now. Automatic pause after at least 50 completed or failed runs in 7 days with a 90% failure rate. The page does not say whether runs can overlap.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/scheduler/jobs_autopilot.go
Locator: scheduler package: jobs_autopilot.go, jobs_issue_wakeup.go, manager.go
Finding: Re-read 2026-10-09: Pinned scheduler/jobs_autopilot.go (8db6cfe): autopilot cron occurrences are planned against database time with the sys_cron_executions lease, on the server and independent of clients.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/service/issue_wakeup_system.go
Locator: issue_wakeup.go, issue_wakeup_condition.go, issue_wakeup_system.go
Finding: Re-read 2026-10-09: Pinned issue_wakeup_system.go and issue_wakeup_condition.go (8db6cfe): wakeup condition types issue_field, children_done, pull_request (checks_finished or merged) and other_issue, with condition state and evidence records.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/releases/tag/v0.5.1
Locator: Release body: MUL-7418 "add issue event and time wakeups (#8467)"
Finding: Re-read 2026-10-09 via the GitHub API: release v0.5.1, published 2026-09-21. The body lists "MUL-7418: add issue event and time wakeups (#8467)".
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/releases/tag/v0.6.0
Locator: Release body: MUL-7680 "issue wakeup v2 - conditions, runaway protection, check-ins and visibility (#8807)"
Finding: Re-read 2026-10-09 via the GitHub API: release v0.6.0, published 2026-09-28. The body lists "MUL-7680: issue wakeup v2 - conditions, runaway protection, check-ins and visibility (#8807)".
Retrieved: 2026-10-09

## multica:COORD-05

Finding: Run history exposes stop and retry; running Claude Code, Codex and Grok tasks can be steered with new instructions; failures and blocked agents surface in the issue and inbox.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Task status change does not automatically stop an active run.
Run-level pause was not found; autopilot pause is a trigger control. Chat versus terminal-mode parity and silent waits were not tested.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cli
Locator: Installation; First connection; Common issue operations; Command reference
Finding: Re-read 2026-10-09: Commands cover issues, projects, labels, agents, skills, autopilots, workspaces, repos, daemon and runtimes. `multica issue cancel-task <task-id> --issue <issue-id>` cancels a run. `multica issue rerun <id>` re-enqueues for the current assignee. Install via curl script, Homebrew or PowerShell. `issue comment update` requires --expected-revision (source, cmd_issue.go).
Retrieved: 2026-10-09

Source: https://multica.ai/docs/autopilots
Locator: Create; Execution mode; Run on a schedule; Run from a webhook; Failures, pausing, and deletion; Permissions
Finding: Re-read 2026-10-09: Autopilots use 5-field cron with IANA time zones, several schedules per autopilot, and run-now. Webhook triggers (JSON up to 256 KiB, idempotency key, GitHub, GitLab and generic event inference, event filters, replay). Create-issue and run-only modes. Run-only failures are not retried and the next schedule still fires. Create-issue waits for an offline runtime; run-only is skipped. Manual pause stops schedules, webhooks and run-now. Automatic pause after at least 50 completed or failed runs in 7 days with a 90% failure rate. The page does not say whether runs can overlap.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/security-model
Locator: The boundary is the daemon user; What Multica does isolate; What is not a boundary
Finding: Re-read 2026-10-09: A run executes with the full permissions of the daemon OS user. Multica does not sandbox the filesystem. Approval prompts are answered automatically. Per-run working directories and per-run API tokens are called conveniences, not a boundary. On the default path Codex runs full-access and Claude Code bypasses permission checks; Codex native sandbox is an optional Windows setting.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/assigning-issues
Locator: Assign and start a run; Assign without starting; Assignment permissions; Use the CLI
Finding: Re-read 2026-10-09: Assignment starts a run unless the issue is in backlog or "Don't start yet" is chosen. Access options: Only me, Specific people, Entire workspace. "a run is bound to its runtime and never moves to another machine." CLI assign commands including --no-start.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/inbox
Locator: What lands in the inbox; Tune notification types
Finding: Re-read 2026-10-09: Inbox covers assignments, updates on subscribed issues, mentions, reactions, run failures on subscribed issues, results of agent-created issues and autopilot notices. Run-failure notices archive once an issue reaches in_review, started, done or closed; blocked keeps them. Banners follow system and browser settings. Mobile push is not covered on this page.
Retrieved: 2026-10-09

## multica:COORD-06

Finding: Issues can be browsed in list, board, table, Gantt and swimlane views with status, assignee and project filters, parent/sub-issue stages and a workspace dashboard; no dependency graph and no fixture-based test.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

No 12-state fixture was run; navigation steps and errors were not measured; no dependency view found.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cli
Locator: Installation; First connection; Common issue operations; Command reference
Finding: Re-read 2026-10-09: Commands cover issues, projects, labels, agents, skills, autopilots, workspaces, repos, daemon and runtimes. `multica issue cancel-task <task-id> --issue <issue-id>` cancels a run. `multica issue rerun <id>` re-enqueues for the current assignee. Install via curl script, Homebrew or PowerShell. `issue comment update` requires --expected-revision (source, cmd_issue.go).
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents
Locator: Agent configuration; Permissions and Access; Online, offline, and workload status
Finding: Re-read 2026-10-09: Agent settings: name, instructions, skills, runtime, model and thinking level, Access (Only me default, Specific people, Entire workspace), concurrency, env, MCP. Availability comes from the runtime and workload from runs. Archiving cancels unfinished runs.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/inbox
Locator: What lands in the inbox; Tune notification types
Finding: Re-read 2026-10-09: Inbox covers assignments, updates on subscribed issues, mentions, reactions, run failures on subscribed issues, results of agent-created issues and autopilot notices. Run-failure notices archive once an issue reaches in_review, started, done or closed; blocked keeps them. Banners follow system and browser settings. Mobile push is not covered on this page.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/views/dashboard/components/dashboard-page.tsx
Locator: KPI cards, daily cost chart, agent run time, failure classes (lines 140-600)
Finding: Re-read 2026-10-09: Pinned source dashboard-page.tsx (8db6cfe): workspace dashboard shows tokens and estimated cost by day and agent, run time, and failure classes; runtime page shows usage per runtime.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/handler/activity.go
Locator: TimelineEntry lines 14-50
Finding: Re-read 2026-10-09: Pinned source internal/handler/activity.go (8db6cfe): the issue timeline interleaves activity and comments with durable actor_type and actor_id attribution; deleted thread roots keep a tombstone.
Retrieved: 2026-10-09

## multica:COORD-07

Finding: No mandatory workflow preset is required; a quick chat or issue works alone, work can grow into sub-issues, and an Agent Builder configures agents from natural language. Per-agent model and thinking level are set on the agent; what ran is visible in the execution log.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Discoverability and auth friction not evaluated.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cloud-quickstart
Locator: Steps 1-5; Common problems
Finding: Re-read 2026-10-09: Sign in by email code or Google on web or Desktop. Connect a computer: Desktop registers it automatically; web uses an install script then multica setup. Create an agent, create an issue and submit. Common problems: offline runtime, "Keep Desktop open".
Retrieved: 2026-10-09

Source: https://multica.ai/docs/squads
Locator: Squad composition; Execution flow after assignment; Leader re-trigger rules
Finding: Re-read 2026-10-09: Assignment to a squad wakes only the leader, which @-mentions members, records an evaluation and stops. The leader is woken again when a delegated member replies, a non-member comments or a member posts progress without a mention. Members can be agents or humans. Provider-native subagents are not described.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents-create
Locator: Starting points; Runtime, model, and thinking level; Configuration after creation; Environment variables and credentials
Finding: Re-read 2026-10-09: Agents start blank or with Build with AI (needs an online runtime). Runtime, model and thinking level. custom_env values are stored in plaintext in the Multica server database; only workspace owners and admins can view or change them. Critical variables such as PATH, HOME and MULTICA_* cannot be overridden.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/chat
Locator: Intro; Multi-turn context; Watch and stop a run
Finding: Re-read 2026-10-09: Private one-to-one chat with an agent outside issues. Attachments supported ("attach images or files"). Session continues where possible. Stopping a run restores unprocessed input to the editor. Offline runtime makes messages wait. No code search, indexing or retrieval statement.
Retrieved: 2026-10-09

## multica:RUN-01

Finding: Connected computers execute local-tool runs while Cloud or self-hosted Multica coordinates them. Vendor-hosted runtime remains waitlisted. A separate Qoder Cloud Agents bridge connects provider-hosted environments.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Multica Cloud/web/desktop; Pinned Multica source
Version: Live docs reread 2026-11-07 through v0.8.1; source at v0.6.1 commit 8db6cfe

Hosted coordination and execution are separate services.
Qoder credentials, remote environment and provider setup are required; this is not a universal Multica-hosted compute claim.
No run, disconnect or self-host deployment test.

Source: https://multica.ai/docs/mobile-app
Locator: Intro; Prerequisites; 7-day limit of free signing
Finding: Current mobile guide says source-built iPhone client, no App Store and no Android, seven-day free signing, server-owned workspace data. Native iPad support is separately documented in release notes; no device test.
Retrieved: 2026-11-07

Source: https://multica.ai/changelog
Locator: v0.8.1 (November 6, 2026), v0.8.0 (November 3) and historical release entries
Finding: Current changelog lists v0.8.1 dated November 6, 2026, after v0.8.0 November 3. Skills revision history, unread filters and Qoder updates are released, not future promises.
Retrieved: 2026-11-07

Source: https://multica.ai/download
Locator: Choose your platform; Prefer the CLI?; Cloud runtime (waitlist); Current version
Finding: Downloads show macOS Apple Silicon/Intel, Windows x64/ARM64 and Linux x64/ARM64 installers; current v0.8.1. Vendor-hosted runtime remains not live, waitlist-only. It is separate from connecting Qoder Cloud Agents.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Connected-computer daemon claims local-tool runs, stores results on server, and has explicit login autostart. Runtimes offline over seven days with no bound agents are cleaned up. Separate Qoder Cloud Agents bridge uses a PAT and remote provider environment without a local daemon; defaults three concurrent runs and 60-minute timeout, configurable.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/desktop-app
Locator: Installation; Built-in daemon; Updating Desktop; Connecting to a self-hosted instance
Finding: Desktop Mac/Windows/Linux shares server data with web, manages separate daemon profiles, auto-updates with OS/architecture feeds and supports self-hosted apiUrl configuration. Terminal issue CLI installed separately.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/cloud-quickstart
Locator: Steps 1-5; Common problems
Finding: Web/Desktop sign-in uses email code or Google. Desktop connects a computer automatically; other computers install CLI and run setup. At least one installed/authenticated coding tool required; no bundled model quota.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/assigning-issues
Locator: Assign and start a run; Assign without starting; Assignment permissions; Use the CLI
Finding: Re-read 2026-10-09: Assignment starts a run unless the issue is in backlog or "Don't start yet" is chosen. Access options: Only me, Specific people, Entire workspace. "a run is bound to its runtime and never moves to another machine." CLI assign commands including --no-start.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/self-host-quickstart
Locator: Multica service and computer; make selfhost; Back up Postgres; Kubernetes
Finding: Re-read 2026-10-09: Self-hosting runs Web, API and PostgreSQL through Docker Compose; Kubernetes via Helm is referenced. No email service by default; codes are printed to backend logs until Resend or SMTP is set. Migrations are forward-only. "Self-hosting replaces only the Multica Cloud part."
Retrieved: 2026-11-07

Source: https://multica.ai/docs/install-agent-runtime
Locator: Choose a tool; Sign in on your machine
Finding: Re-read 2026-10-09: Sign in to each tool on its own machine; Multica never receives login tokens. "Supported" means Multica can invoke the CLI; it does not supply the tool account, subscription or model quota. The table lists 25 CLIs. Minimum versions for eight tools; below minimum the runtime is not registered.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/how-multica-works
Locator: The full path of a run; The data and execution boundary
Finding: Re-read 2026-10-09: The server coordinates and records work; connected computers execute it. Code directories and local files stay on the connected computer. The server stores issues, comments, statuses, run records and results. Custom env values are stored on the server. The boundary is the same on Cloud and self-hosted.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/core/types/billing.ts
Locator: Header comment lines 1-12; BillingBatch; BillingTopup
Finding: Re-read 2026-10-09: Pinned billing types (8db6cfe): credit wallet with 1 USD = 1000 credits, credit batches and top-up states. Header states a separate multica-cloud billing service. No public description of what credits purchase.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/SELF_HOSTING.md
Locator: Telemetry paragraph line 19; Kubernetes section
Finding: Re-read 2026-10-09: Pinned SELF_HOSTING.md (8db6cfe): configure RESEND_API_KEY for email codes; without email the code is printed to backend logs; fixed code only when APP_ENV=development.
Retrieved: 2026-11-07

## multica:RUN-02

Finding: GitHub repository runs get their own worktree from a cached bare clone; local git directories can use per-run worktrees on agent/<agent>/<issue> branches, while a direct local-directory run uses the user's own checkout in place unless Parallel mode is chosen. Multica states it provides no filesystem, process or network sandbox and recommends a dedicated user, container or VM.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2027-01-07
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Default daemon path; Windows opt-in Codex sandbox is the one documented exception.
Two-change code separation was not tested.

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/security-model
Locator: The boundary is the daemon user; What Multica does isolate; What is not a boundary
Finding: Re-read 2026-10-09: A run executes with the full permissions of the daemon OS user. Multica does not sandbox the filesystem. Approval prompts are answered automatically. Per-run working directories and per-run API tokens are called conveniences, not a boundary. On the default path Codex runs full-access and Claude Code bypasses permission checks; Codex native sandbox is an optional Windows setting.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents-create
Locator: Starting points; Runtime, model, and thinking level; Configuration after creation; Environment variables and credentials
Finding: Re-read 2026-10-09: Agents start blank or with Build with AI (needs an online runtime). Runtime, model and thinking level. custom_env values are stored in plaintext in the Multica server database; only workspace owners and admins can view or change them. Critical variables such as PATH, HOME and MULTICA_* cannot be overridden.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/project-resources
Locator: Add a GitHub repository; Add a local directory; How runs share the directory
Finding: Re-read 2026-10-09: GitHub repositories are checked out per run through worktrees in a runtime-managed workspace. Local git directories run in place (serialized, directory lock) or in Parallel worktree mode on agent/<agent>/<issue> branches. Non-git directories run serially. Dependency caches, snapshots and port allocation are not mentioned.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/daemon/repocache/cache.go
Locator: package header lines 1-35, CachedRepo line 160
Finding: Re-read 2026-10-09: Pinned source repocache/cache.go (8db6cfe): the daemon keeps cached bare clones and creates per-task git worktrees from them. The file also calls isolateWorktreeIdentityContext for each worktree checkout and reconciles a Co-authored-by hook.
Retrieved: 2026-10-09

## multica:RUN-03

Finding: Transient failures auto-retry (default 2 attempts, 3 for tool network errors) and a daemon restart reclaims interrupted runs; retries keep files and continue the session where possible. Credential, quota and configuration errors need a manual fix and retry.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Recovery and duplicate-action safety not observed.
Compute replacement for a multi-day epic is not supported by moving runs between machines.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/providers
Locator: Supported tools table; Model sources; Session resumption; MCP configuration; Skill injection
Finding: Re-read 2026-10-09: Supported tools table (25 CLIs). Model lists come from each runtime. QwenPaw and MiniMax Code do not accept a model chosen in Multica. No full provider list. Session resumption continues for tools marked as supporting it; MiniMax Code starts a fresh session. MCP is not read for Antigravity, GitHub Copilot CLI, DevEco Code or Pi.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/desktop-app
Locator: Installation; Built-in daemon; Updating Desktop; Connecting to a self-hosted instance
Finding: Re-read 2026-10-09: Installers: macOS (.dmg), Windows (.exe), Linux (.AppImage, .deb, .rpm). Desktop starts a built-in daemon after sign-in. Auto-update on by default. Self-hosted connection through desktop.json. No minimum OS version stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cloud-quickstart
Locator: Steps 1-5; Common problems
Finding: Re-read 2026-10-09: Sign in by email code or Google on web or Desktop. Connect a computer: Desktop registers it automatically; web uses an install script then multica setup. Create an agent, create an issue and submit. Common problems: offline runtime, "Keep Desktop open".
Retrieved: 2026-10-09

Source: https://multica.ai/docs/assigning-issues
Locator: Assign and start a run; Assign without starting; Assignment permissions; Use the CLI
Finding: Re-read 2026-10-09: Assignment starts a run unless the issue is in backlog or "Don't start yet" is chosen. Access options: Only me, Specific people, Entire workspace. "a run is bound to its runtime and never moves to another machine." CLI assign commands including --no-start.
Retrieved: 2026-10-09

## multica:RUN-04

Finding: Repository checkouts use cached bare clones and per-task worktrees; the user supplies and maintains tool installs and environments. No product-managed dependency caches, snapshots, service or port allocation were found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Dependency installs, ports, collisions, snapshots and services were not found as product features.

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cloud-quickstart
Locator: Steps 1-5; Common problems
Finding: Re-read 2026-10-09: Sign in by email code or Google on web or Desktop. Connect a computer: Desktop registers it automatically; web uses an install script then multica setup. Create an agent, create an issue and submit. Common problems: offline runtime, "Keep Desktop open".
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents-create
Locator: Starting points; Runtime, model, and thinking level; Configuration after creation; Environment variables and credentials
Finding: Re-read 2026-10-09: Agents start blank or with Build with AI (needs an online runtime). Runtime, model and thinking level. custom_env values are stored in plaintext in the Multica server database; only workspace owners and admins can view or change them. Critical variables such as PATH, HOME and MULTICA_* cannot be overridden.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/daemon/repocache/cache.go
Locator: package header lines 1-35, CachedRepo line 160
Finding: Re-read 2026-10-09: Pinned source repocache/cache.go (8db6cfe): the daemon keeps cached bare clones and creates per-task git worktrees from them. The file also calls isolateWorktreeIdentityContext for each worktree checkout and reconciles a Co-authored-by hook.
Retrieved: 2026-10-09

## multica:RUN-05

Finding: The daemon enforces a configurable absolute run time limit (default none), an idle watchdog and a tool watchdog (default 2h silence), and classifies provider failures. The platform reports an iteration_limit failure reason when a run reaches its per-run iteration cap. Token and spend budgets were not established in the sources read.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Token and spend limits not found; spending limit requested in open issue #6950.
Iteration cap: iteration_limit is a platform failure reason (failure.go lines 92 to 96; tasks.mdx line 102). The cap value is not stated, and how an operator sets it was not established.
Claude Code runs receive --max-turns from the run MaxTurns option in source (claude.go line 1110). How an operator sets that option was not established.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/environment-variables
Locator: Daemon configuration; Observability and analytics; Server-side LLM
Finding: Re-read 2026-10-09: Telemetry on by default: one anonymous self-host snapshot per UTC day; DO_NOT_TRACK=1 or true disables it; ANALYTICS_DISABLED=true turns off PostHog. MULTICA_DAEMON_MAX_CONCURRENT_TASKS default 20. MULTICA_AGENT_TIMEOUT default 0 (no limit). MULTICA_AGENT_IDLE_WATCHDOG default 2h; tool watchdog same by default. No iteration limit variable listed. Optional server-side LLM for chat titles sends chat content to a configured endpoint.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/pkg/agent/claude.go
Locator: Execute (lines 36-90); buildClaudeArgs lines 1060-1115
Finding: Re-read 2026-10-09: Pinned claude.go (8db6cfe): runs Claude Code with stream-json input and output and keeps stdin open after the first message. Passes --max-turns from the run MaxTurns option (line 1110).
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/issues
Locator: Searches 2026-10-08: Android, App Store, SSO/SAML, dependency/blocked-by, budget/spend limit, cloud runtime
Finding: Re-read 2026-10-09: GitHub searches on 2026-10-09 return 16 open Android results, 12 App Store results and 13 SSO results. Open community requests: #3864 and #3412 (Android), #6996 (mobile push), #7794 and #1218 (blocker dependencies), #6950 (spending limit), #5423 (Azure AAD), #4151 (open-source the hosted runtime). None is a vendor announcement. The repository reports 1,839 open issues and pull requests combined.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/pkg/taskfailure/failure.go
Locator: Reason constants: ReasonIterationLimit (lines 92 to 96)
Finding: Re-read 2026-10-09: ReasonIterationLimit = "iteration_limit" is a platform-side failure reason. The comment says the agent reached its per-run iteration cap and emitted a fallback message; it is treated as a Multica-imposed budget. The cap value is not in this file.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/apps/docs/content/docs/tasks.mdx
Locator: Failure reasons table, row iteration_limit (line 102)
Finding: Re-read 2026-10-09: the failure reasons table lists iteration_limit, "Hit the iteration cap for a run", with the action "Narrow the issue scope". The table does not give the cap value or how an operator sets it.
Retrieved: 2026-10-09

## multica:RUN-06

Finding: Execution log exposes run state, timing and streaming transcript with classified failure reasons; a workspace dashboard shows usage, estimated cost, run time and failures by day and agent, and the CLI reports issue and runtime usage.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Displayed completion and reconnect consistency not independently tested; traces not found.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cli
Locator: Installation; First connection; Common issue operations; Command reference
Finding: Re-read 2026-10-09: Commands cover issues, projects, labels, agents, skills, autopilots, workspaces, repos, daemon and runtimes. `multica issue cancel-task <task-id> --issue <issue-id>` cancels a run. `multica issue rerun <id>` re-enqueues for the current assignee. Install via curl script, Homebrew or PowerShell. `issue comment update` requires --expected-revision (source, cmd_issue.go).
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/views/dashboard/components/dashboard-page.tsx
Locator: KPI cards, daily cost chart, agent run time, failure classes (lines 140-600)
Finding: Re-read 2026-10-09: Pinned source dashboard-page.tsx (8db6cfe): workspace dashboard shows tokens and estimated cost by day and agent, run time, and failure classes; runtime page shows usage per runtime.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/views/issues/components/execution-log-section.tsx
Locator: cumulativeCostAt, per-run usage cost, sparkline (lines 28-420)
Finding: Re-read 2026-10-09: Pinned source execution-log-section.tsx (8db6cfe): issue execution log shows per-run usage and cost and a cumulative cost curve estimated from tokens.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/views/dashboard/components/dashboard-page.tsx
Locator: lines 131 to 244 (usage dashboard; six rollup queries)
Finding: Re-read 2026-10-09: the project filter (projectId) is passed to all six rollups: daily usage, by-agent usage, run-time, run-time daily, failures daily and failures by agent (lines 206 to 232). Source-inspected only; no hands-on parity check.
Retrieved: 2026-10-09

## multica:RUN-07

Finding: Estimated per-run and per-issue cost appears in the execution log with a cumulative curve, and the CLI reports issue usage. Figures are the provider-reported cost plus a rate-table estimate for uncosted tokens; custom rates apply only to the estimated part (packages/views/runtimes/utils.ts lines 676 to 702 at 8db6cfe). Billed invoice reconciliation, subscription allowance and compute fees are not established.

Verification: sourced
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Pinned Multica source; Multica Cloud/web/desktop
Version: commit 8db6cfe; v0.6.1

Cost semantics, estimates and provider reconciliation unverified.
Compute fees, alerts and export not found.

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/cmd/multica/cmd_issue.go
Locator: issueCmd subcommands lines 222-456 (list/get/pull-requests/children/create/update/assign/status/comment/runs/usage/rerun/cancel-task/search); comment update --expected-revision lines 725 and 2584-2613
Finding: Re-read 2026-10-09: Pinned source cmd_issue.go (8db6cfe): issue subcommands include list, get, pull-requests, children, create, update, assign, status, comment, runs, run-messages, usage, rerun, cancel-task and search; comment update requires --expected-revision; calls go through the shared server API.
Retrieved: 2026-11-07

Source: https://multica.ai/changelog
Locator: v0.8.1 (November 6, 2026), v0.8.0 (November 3) and historical release entries
Finding: Current changelog lists v0.8.1 dated November 6, 2026, after v0.8.0 November 3. Skills revision history, unread filters and Qoder updates are released, not future promises.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/views/runtimes/utils.ts
Locator: MODEL_PRICING table lines 203-400; estimateCost
Finding: Re-read 2026-10-09: Pinned runtimes/utils.ts (8db6cfe): static MODEL_PRICING table of provider rates; estimateCost computes cost from token counts and can prefer per-turn prices.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/views/dashboard/components/dashboard-page.tsx
Locator: KPI cards, daily cost chart, agent run time, failure classes (lines 140-600)
Finding: Re-read 2026-10-09: Pinned source dashboard-page.tsx (8db6cfe): workspace dashboard shows tokens and estimated cost by day and agent, run time, and failure classes; runtime page shows usage per runtime.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/views/issues/components/execution-log-section.tsx
Locator: cumulativeCostAt, per-run usage cost, sparkline (lines 28-420)
Finding: Re-read 2026-10-09: Pinned source execution-log-section.tsx (8db6cfe): issue execution log shows per-run usage and cost and a cumulative cost curve estimated from tokens.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/views/runtimes/utils.ts
Locator: lines 676 to 702, estimateCost
Finding: Re-read 2026-10-09: estimateCost adds the provider-reported cost_usd_ticks (authoritative half) to a rate-table estimate for uncosted tokens; custom pricing overrides apply only to the estimated half.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/analytics
Locator: Page layout; Cost calculation; Issue cost; From the CLI
Finding: Usage rolls up every five minutes and supports project, runtime, model, agent and issue detail. Costs are estimated unless the runtime reports its own charge; unpriced models can count as zero cost.
Retrieved: 2026-11-07

## multica:RUN-08

Finding: A daemon-wide cap (20) and per-agent cap (6) queue excess runs, and a directory lock serializes runs on the same local directory; CPU/RAM-aware scheduling and check de-duplication were not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Fairness, recovery under load, check reuse and bounded browser-test queues were not tested.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/project-resources
Locator: Add a GitHub repository; Add a local directory; How runs share the directory
Finding: Re-read 2026-10-09: GitHub repositories are checked out per run through worktrees in a runtime-managed workspace. Local git directories run in place (serialized, directory lock) or in Parallel worktree mode on agent/<agent>/<issue> branches. Non-git directories run serially. Dependency caches, snapshots and port allocation are not mentioned.
Retrieved: 2026-10-09

## multica:TEAM-01

Finding: Workspace members share issues and comments with attribution; members can each comment, mention and steer agents with coalescing of simultaneous instructions; presence indicators and plan restrictions were not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Presence and plan restrictions not found.

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/cmd/multica/cmd_issue.go
Locator: issueCmd subcommands lines 222-456 (list/get/pull-requests/children/create/update/assign/status/comment/runs/usage/rerun/cancel-task/search); comment update --expected-revision lines 725 and 2584-2613
Finding: Re-read 2026-10-09: Pinned source cmd_issue.go (8db6cfe): issue subcommands include list, get, pull-requests, children, create, update, assign, status, comment, runs, run-messages, usage, rerun, cancel-task and search; comment update requires --expected-revision; calls go through the shared server API.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/members-roles
Locator: The three roles; Invite members; Remove a member
Finding: Re-read 2026-10-09: Roles owner, admin, member. Invitations valid 7 days. "A removed member loses workspace access immediately"; their runtimes are deactivated, agents archived and unfinished runs cancelled. Roles do not decide who can run an agent. No SSO or provisioning mention.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents
Locator: Agent configuration; Permissions and Access; Online, offline, and workload status
Finding: Re-read 2026-10-09: Agent settings: name, instructions, skills, runtime, model and thinking level, Access (Only me default, Specific people, Entire workspace), concurrency, env, MCP. Availability comes from the runtime and workload from runs. Archiving cancels unfinished runs.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/mentioning-agents
Locator: Trigger preview; Automatic routing of replies; Coalescing consecutive comments; Access restrictions
Finding: Re-read 2026-10-09: Mentions and replies start runs; a trigger preview is shown before sending. Simultaneous mentions of different agents are not merged; consecutive comments to one agent merge into the pending run. Runs wait when the runtime is offline. Access settings apply to mentions. No presence indicators.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/chat
Locator: Intro; Multi-turn context; Watch and stop a run
Finding: Re-read 2026-10-09: Private one-to-one chat with an agent outside issues. Attachments supported ("attach images or files"). Session continues where possible. Stopping a run restores unprocessed input to the editor. Offline runtime makes messages wait. No code search, indexing or retrieval statement.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/comments
Locator: Post comments; Annotate; Resolve a discussion; @-mentions; Edit and delete
Finding: Re-read 2026-10-09: Threaded comments with attachments, annotations, resolve or conclusion, mentions and reactions. Authors edit or delete their own comments; owners and admins edit or delete anyone's; agent-posted comments can only be deleted. Deletion is permanent. Drafts expire after 30 days. Retention and export not stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/workspaces
Locator: What a workspace contains; Delete a workspace
Finding: Re-read 2026-10-09: A workspace is the isolation boundary for members, issues, projects, agents, skills and run history. Owners and admins manage settings; members view them. Deletion is permanent. No export or portability statement. No SSO mention.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/handler/activity.go
Locator: TimelineEntry lines 14-50
Finding: Re-read 2026-10-09: Pinned source internal/handler/activity.go (8db6cfe): the issue timeline interleaves activity and comments with durable actor_type and actor_id attribution; deleted thread roots keep a tombstone.
Retrieved: 2026-10-09

## multica:TEAM-02

Finding: Workspaces have owner, admin and member roles, 7-day invitations and immediate revocation on removal; agent Access (only me, specific people, entire workspace) governs who can run an agent. SSO and provisioning were not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Project and task roles, SSO and provisioning were not found; SSO requests are open (#5423, #1392).
Plan scope of roles unknown.

Source: https://multica.ai/docs/members-roles
Locator: The three roles; Invite members; Remove a member
Finding: Re-read 2026-10-09: Roles owner, admin, member. Invitations valid 7 days. "A removed member loses workspace access immediately"; their runtimes are deactivated, agents archived and unfinished runs cancelled. Roles do not decide who can run an agent. No SSO or provisioning mention.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents
Locator: Agent configuration; Permissions and Access; Online, offline, and workload status
Finding: Re-read 2026-10-09: Agent settings: name, instructions, skills, runtime, model and thinking level, Access (Only me default, Specific people, Entire workspace), concurrency, env, MCP. Availability comes from the runtime and workload from runs. Archiving cancels unfinished runs.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/auth-tokens
Locator: Personal access tokens; Using a PAT in API requests; Temporary tokens for agent runs
Finding: Re-read 2026-10-09: Browser sessions last 30 days. Personal access tokens (mul_ prefix, 30 days, 90 days, 1 year or never) work against REST endpoints. multica login creates a 90-day token. Run-scoped mat_ tokens last at most 24 hours and attribute actions to the agent. No dedicated REST API reference.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/workspaces
Locator: What a workspace contains; Delete a workspace
Finding: Re-read 2026-10-09: A workspace is the isolation boundary for members, issues, projects, agents, skills and run history. Owners and admins manage settings; members view them. Deletion is permanent. No export or portability statement. No SSO mention.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/issues
Locator: Searches 2026-10-08: Android, App Store, SSO/SAML, dependency/blocked-by, budget/spend limit, cloud runtime
Finding: Re-read 2026-10-09: GitHub searches on 2026-10-09 return 16 open Android results, 12 App Store results and 13 SSO results. Open community requests: #3864 and #3412 (Android), #6996 (mobile push), #7794 and #1218 (blocker dependencies), #6950 (spending limit), #5423 (Azure AAD), #4151 (open-source the hosted runtime). None is a vendor announcement. The repository reports 1,839 open issues and pull requests combined.
Retrieved: 2026-10-09

## multica:TEAM-03

Finding: Runs are unsandboxed and tool approval prompts are answered automatically by default; agent Access controls who can trigger agents, but no per-command, tool, network or delivery approval gate was found.

Verification: sourced
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Delivery, secret and tunnel approvals, approver identity and allow/deny/revoke flows were not found.
Scope is the default daemon path; Windows Codex opt-in is an exception.

Source: https://multica.ai/docs/security-model
Locator: The boundary is the daemon user; What Multica does isolate; What is not a boundary
Finding: Re-read 2026-10-09: A run executes with the full permissions of the daemon OS user. Multica does not sandbox the filesystem. Approval prompts are answered automatically. Per-run working directories and per-run API tokens are called conveniences, not a boundary. On the default path Codex runs full-access and Claude Code bypasses permission checks; Codex native sandbox is an optional Windows setting.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents
Locator: Agent configuration; Permissions and Access; Online, offline, and workload status
Finding: Re-read 2026-10-09: Agent settings: name, instructions, skills, runtime, model and thinking level, Access (Only me default, Specific people, Entire workspace), concurrency, env, MCP. Availability comes from the runtime and workload from runs. Archiving cancels unfinished runs.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/auth-tokens
Locator: Personal access tokens; Using a PAT in API requests; Temporary tokens for agent runs
Finding: Re-read 2026-10-09: Browser sessions last 30 days. Personal access tokens (mul_ prefix, 30 days, 90 days, 1 year or never) work against REST endpoints. multica login creates a 90-day token. Run-scoped mat_ tokens last at most 24 hours and attribute actions to the agent. No dedicated REST API reference.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/CLI_AND_DAEMON.md
Locator: Supported Agents table lines 217-250; How It Works lines 251-258; per-tool launch notes lines 355-396
Finding: Re-read 2026-10-09: Pinned CLI_AND_DAEMON.md (8db6cfe): runtime table with protocol families including ACP for several tools; MCP servers passed over ACP session/new and resume requests; MiniMax Code 0.1.2 has no session-loading capability; remote MCP transports depend on declared capability.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/pkg/agent/claude.go
Locator: Execute (lines 36-90); buildClaudeArgs lines 1060-1115
Finding: Re-read 2026-10-09: Pinned claude.go (8db6cfe): runs Claude Code with stream-json input and output and keeps stdin open after the first message. Passes --max-turns from the run MaxTurns option (line 1110).
Retrieved: 2026-10-09

Source: https://multica.ai/docs/security-model
Locator: Recommended setups; What Multica does isolate
Finding: The daemon normally runs with its OS user’s full permissions. Operators can put it in a dedicated account, container or VM; these are externally configured boundaries.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/auth-tokens
Locator: Temporary tokens for agent runs
Finding: A temporary run token is bound to user, workspace, agent and run, lasts at most 24 hours and is cleaned up at run end.
Retrieved: 2026-11-07

## multica:TEAM-04

Finding: The issue timeline records attributed activity and comments with actor type and id, comment revisions and resolve state; history is server-stored, but comments can be edited or deleted by authors and admins. No export or integrity model was found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Approvals, export and integrity not found; do not call this an audit log. Reads and changes of agent env values leave an audit record (agent docs).

Source: https://multica.ai/privacy
Locator: Last updated September 24, 2026; AI features; Who we share information with; Where information is stored; How long we keep information
Finding: Re-read 2026-10-09: Last updated 2026-09-24. Operated by Index Labs (Hong Kong) Limited. Hosted on AWS and Vercel. Processors: AWS, Vercel, Stripe, Resend, PostHog, Google sign-in and LLM providers. "Multica does not use your content to train AI models." Cloud chat features send messages to a third-party LLM provider. Privacy requests answered "within 30 days". support@multica.ai is listed for privacy questions and requests. No export of workspace data on exit is stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents-create
Locator: Starting points; Runtime, model, and thinking level; Configuration after creation; Environment variables and credentials
Finding: Re-read 2026-10-09: Agents start blank or with Build with AI (needs an online runtime). Runtime, model and thinking level. custom_env values are stored in plaintext in the Multica server database; only workspace owners and admins can view or change them. Critical variables such as PATH, HOME and MULTICA_* cannot be overridden.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/comments
Locator: Post comments; Annotate; Resolve a discussion; @-mentions; Edit and delete
Finding: Re-read 2026-10-09: Threaded comments with attachments, annotations, resolve or conclusion, mentions and reactions. Authors edit or delete their own comments; owners and admins edit or delete anyone's; agent-posted comments can only be deleted. Deletion is permanent. Drafts expire after 30 days. Retention and export not stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/workspaces
Locator: What a workspace contains; Delete a workspace
Finding: Re-read 2026-10-09: A workspace is the isolation boundary for members, issues, projects, agents, skills and run history. Owners and admins manage settings; members view them. Deletion is permanent. No export or portability statement. No SSO mention.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/handler/activity.go
Locator: TimelineEntry lines 14-50
Finding: Re-read 2026-10-09: Pinned source internal/handler/activity.go (8db6cfe): the issue timeline interleaves activity and comments with durable actor_type and actor_id attribution; deleted thread roots keep a tombstone.
Retrieved: 2026-10-09

## multica:TEAM-05

Finding: Review is issue- and PR-centric: in_review status, a pull request card with state, size, CI and mergeability, reviewer agents commenting on the issue, and a deliverables sidebar; the diff itself is reviewed on GitHub.

Verification: sourced
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Keyboard and narrow-screen use, effort and missed evidence were not measured; request-changes rests on documented workflow send-back and comments, not a test.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/inbox
Locator: What lands in the inbox; Tune notification types
Finding: Re-read 2026-10-09: Inbox covers assignments, updates on subscribed issues, mentions, reactions, run failures on subscribed issues, results of agent-created issues and autopilot notices. Run-failure notices archive once an issue reaches in_review, started, done or closed; blocked keeps them. Banners follow system and browser settings. Mobile push is not covered on this page.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/comments
Locator: Post comments; Annotate; Resolve a discussion; @-mentions; Edit and delete
Finding: Re-read 2026-10-09: Threaded comments with attachments, annotations, resolve or conclusion, mentions and reactions. Authors edit or delete their own comments; owners and admins edit or delete anyone's; agent-posted comments can only be deleted. Deletion is permanent. Drafts expire after 30 days. Retention and export not stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tutorial
Locator: Prepare the site repository; Reviewer; Merge and close out; Form a squad; Automation
Finding: Re-read 2026-10-09: walkthrough. The first agent uses Claude with Sonnet (line 60); the Engineer agent, created through Multica helper, uses Opus (line 85); Reviewer uses Codex with GPT-5.6 Sol in a fresh context and never edits code (line 208). The Engineer opens the PR using the GitHub CLI on the user's computer (line 95). Merging is requested in a comment. Merge-driven status is configurable: the GitHub integration docs say that after PRs merge the issue moves to Done by default or is left alone (github-integration, "After PRs merge, move the issue to"). A PR that references an issue identifier links automatically (line 107).
Retrieved: 2026-10-09

Source: https://multica.ai/docs/workflows
Locator: Agent steps; Human steps; Handoffs
Finding: Workflow columns can hand issues to an agent or person, define the next state after completion and a send-back path. Blocked is reserved for outside obstacles.
Retrieved: 2026-11-07

## multica:TEAM-06

Finding: Issues, comments, transcripts and attachments live on the server, so another member can follow the same issue and reassign it; runs stay bound to the original runtime. A private runtime limits who can create agents on that computer, and an agent's Access setting controls who can run that agent (daemon-runtimes "Private and public runtimes"; agents "Access").

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

No takeover test; context loss and harness transfer were not evaluated.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/members-roles
Locator: The three roles; Invite members; Remove a member
Finding: Re-read 2026-10-09: Roles owner, admin, member. Invitations valid 7 days. "A removed member loses workspace access immediately"; their runtimes are deactivated, agents archived and unfinished runs cancelled. Roles do not decide who can run an agent. No SSO or provisioning mention.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents
Locator: Agent configuration; Permissions and Access; Online, offline, and workload status
Finding: Re-read 2026-10-09: Agent settings: name, instructions, skills, runtime, model and thinking level, Access (Only me default, Specific people, Entire workspace), concurrency, env, MCP. Availability comes from the runtime and workload from runs. Archiving cancels unfinished runs.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/assigning-issues
Locator: Assign and start a run; Assign without starting; Assignment permissions; Use the CLI
Finding: Re-read 2026-10-09: Assignment starts a run unless the issue is in backlog or "Don't start yet" is chosen. Access options: Only me, Specific people, Entire workspace. "a run is bound to its runtime and never moves to another machine." CLI assign commands including --no-start.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/handler/activity.go
Locator: TimelineEntry lines 14-50
Finding: Re-read 2026-10-09: Pinned source internal/handler/activity.go (8db6cfe): the issue timeline interleaves activity and comments with durable actor_type and actor_id attribution; deleted thread roots keep a tombstone.
Retrieved: 2026-10-09

## multica:SHIP-01

Finding: Issues retain comments, attachments with versions, a deliverables sidebar and in-place previews (HTML, Markdown, CSV, JSON, YAML, Mermaid, images), run transcripts and linked pull requests; diffs, test outputs and recordings are reached through GitHub or run transcripts.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Recordings and prompt-leakage checks not found; revision-exact links not established.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/squads
Locator: Squad composition; Execution flow after assignment; Leader re-trigger rules
Finding: Re-read 2026-10-09: Assignment to a squad wakes only the leader, which @-mentions members, records an evaluation and stops. The leader is woken again when a delegated member replies, a non-member comments or a member posts progress without a mention. Members can be agents or humans. Provider-native subagents are not described.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/mentioning-agents
Locator: Trigger preview; Automatic routing of replies; Coalescing consecutive comments; Access restrictions
Finding: Re-read 2026-10-09: Mentions and replies start runs; a trigger preview is shown before sending. Simultaneous mentions of different agents are not merged; consecutive comments to one agent merge into the pending run. Runs wait when the runtime is offline. Access settings apply to mentions. No presence indicators.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/comments
Locator: Post comments; Annotate; Resolve a discussion; @-mentions; Edit and delete
Finding: Re-read 2026-10-09: Threaded comments with attachments, annotations, resolve or conclusion, mentions and reactions. Authors edit or delete their own comments; owners and admins edit or delete anyone's; agent-posted comments can only be deleted. Deletion is permanent. Drafts expire after 30 days. Retention and export not stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/workspaces
Locator: What a workspace contains; Delete a workspace
Finding: Re-read 2026-10-09: A workspace is the isolation boundary for members, issues, projects, agents, skills and run history. Owners and admins manage settings; members view them. Deletion is permanent. No export or portability statement. No SSO mention.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tutorial
Locator: Prepare the site repository; Reviewer; Merge and close out; Form a squad; Automation
Finding: Re-read 2026-10-09: walkthrough. The first agent uses Claude with Sonnet (line 60); the Engineer agent, created through Multica helper, uses Opus (line 85); Reviewer uses Codex with GPT-5.6 Sol in a fresh context and never edits code (line 208). The Engineer opens the PR using the GitHub CLI on the user's computer (line 95). Merging is requested in a comment. Merge-driven status is configurable: the GitHub integration docs say that after PRs merge the issue moves to Done by default or is left alone (github-integration, "After PRs merge, move the issue to"). A PR that references an issue identifier links automatically (line 107).
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/handler/activity.go
Locator: TimelineEntry lines 14-50
Finding: Re-read 2026-10-09: Pinned source internal/handler/activity.go (8db6cfe): the issue timeline interleaves activity and comments with durable actor_type and actor_id attribution; deleted thread roots keep a tombstone.
Retrieved: 2026-10-09

## multica:SHIP-02

Finding: Multica shows CI status and mergeability on the PR card and can move an issue when linked PRs merge, but no product-enforced test, lint, review or spec gate was found; checks run by agents or CI are not a Multica gate.

Verification: sourced
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Failure blocking, required checks and spec conformance not found.

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tutorial
Locator: Prepare the site repository; Reviewer; Merge and close out; Form a squad; Automation
Finding: Re-read 2026-10-09: walkthrough. The first agent uses Claude with Sonnet (line 60); the Engineer agent, created through Multica helper, uses Opus (line 85); Reviewer uses Codex with GPT-5.6 Sol in a fresh context and never edits code (line 208). The Engineer opens the PR using the GitHub CLI on the user's computer (line 95). Merging is requested in a comment. Merge-driven status is configurable: the GitHub integration docs say that after PRs merge the issue moves to Done by default or is left alone (github-integration, "After PRs merge, move the issue to"). A PR that references an issue identifier links automatically (line 107).
Retrieved: 2026-10-09

## multica:SHIP-03

Finding: Agents run Git and the GitHub CLI themselves; Multica adds per-task worktrees, a task Git identity, optional Co-authored-by trailers, PR auto-linking, CI and mergeability display and merge-driven status changes.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

No draft PR was completed in a fixture; rebase and CI response flows not found.
The task Git identity is source-inspected only; the docs do not describe it.

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/project-resources
Locator: Add a GitHub repository; Add a local directory; How runs share the directory
Finding: Re-read 2026-10-09: GitHub repositories are checked out per run through worktrees in a runtime-managed workspace. Local git directories run in place (serialized, directory lock) or in Parallel worktree mode on agent/<agent>/<issue> branches. Non-git directories run serially. Dependency caches, snapshots and port allocation are not mentioned.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tutorial
Locator: Prepare the site repository; Reviewer; Merge and close out; Form a squad; Automation
Finding: Re-read 2026-10-09: walkthrough. The first agent uses Claude with Sonnet (line 60); the Engineer agent, created through Multica helper, uses Opus (line 85); Reviewer uses Codex with GPT-5.6 Sol in a fresh context and never edits code (line 208). The Engineer opens the PR using the GitHub CLI on the user's computer (line 95). Merging is requested in a comment. Merge-driven status is configurable: the GitHub integration docs say that after PRs merge the issue moves to Done by default or is left alone (github-integration, "After PRs merge, move the issue to"). A PR that references an issue identifier links automatically (line 107).
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/daemon/repocache/cache.go
Locator: package header lines 1-35, CachedRepo line 160
Finding: Re-read 2026-10-09: Pinned source repocache/cache.go (8db6cfe): the daemon keeps cached bare clones and creates per-task git worktrees from them. The file also calls isolateWorktreeIdentityContext for each worktree checkout and reconciles a Co-authored-by hook.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/server/internal/daemon/repocache/identity.go
Locator: isolateWorktreeIdentityContext definition, lines 15 to 26; masks inherited identity and writes multica-identity.config
Finding: Re-read 2026-10-09: isolateWorktreeIdentityContext (line 26) masks identity inherited from the shared cache for each worktree; cache.go calls it at lines 892, 916 and 946.
Retrieved: 2026-10-09

## multica:SHIP-04

Finding: Merging is done by humans or by agents on request through the user's own Git tooling; the GitHub App is read-only. Release and deploy authority and exact-revision checks were not found.

Verification: sourced
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Approval, release, deploy and exact-revision checks not found.

Source: https://multica.ai/docs/security-model
Locator: The boundary is the daemon user; What Multica does isolate; What is not a boundary
Finding: Re-read 2026-10-09: A run executes with the full permissions of the daemon OS user. Multica does not sandbox the filesystem. Approval prompts are answered automatically. Per-run working directories and per-run API tokens are called conveniences, not a boundary. On the default path Codex runs full-access and Claude Code bypasses permission checks; Codex native sandbox is an optional Windows setting.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents
Locator: Agent configuration; Permissions and Access; Online, offline, and workload status
Finding: Re-read 2026-10-09: Agent settings: name, instructions, skills, runtime, model and thinking level, Access (Only me default, Specific people, Entire workspace), concurrency, env, MCP. Availability comes from the runtime and workload from runs. Archiving cancels unfinished runs.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/tutorial
Locator: Prepare the site repository; Reviewer; Merge and close out; Form a squad; Automation
Finding: Re-read 2026-10-09: walkthrough. The first agent uses Claude with Sonnet (line 60); the Engineer agent, created through Multica helper, uses Opus (line 85); Reviewer uses Codex with GPT-5.6 Sol in a fresh context and never edits code (line 208). The Engineer opens the PR using the GitHub CLI on the user's computer (line 95). Merging is requested in a comment. Merge-driven status is configurable: the GitHub integration docs say that after PRs merge the issue moves to Done by default or is left alone (github-integration, "After PRs merge, move the issue to"). A PR that references an issue identifier links automatically (line 107).
Retrieved: 2026-10-09

## multica:SHIP-05

Finding: Linked pull requests show state and merge status on the issue and merged PRs can complete it; no build, deployment, rollback or stale-approval invalidation links were found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Build, deployment, rollback and revert links were not found.

Source: https://multica.ai/docs/tasks
Locator: Execution lifecycle; Viewing run history; Failures and automatic retries; Manual retry; State and timeout quick reference
Finding: Re-read 2026-10-09: States deferred, queued, dispatched, waiting_local_directory, running, completed, failed, cancelled. Runs can be stopped from the execution log while deferred, queued, starting, waiting or running. A status change does not stop an active run. Retry from the execution log invokes the original agent and keeps files and session when possible. No pause control. Failure reasons include agent_blocked and missing_config. Dispatched runs fail after 5 minutes. Heartbeat every 15 seconds. Automatic retries only for transient faults (2 attempts, 3 for network interruptions). Transcripts kept.
Retrieved: 2026-10-09

Source: https://multica.ai/changelog
Locator: Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01); v0.5.1, v0.5.2, v0.6.0, v0.6.1 entries; v0.4.35, v0.4.34
Finding: Re-read 2026-10-09: Re-read first 100,000 of 188,543 characters (remainder not read). Releases v0.1.0 (2026-03-22) to v0.6.1 (2026-10-01). Dated entries cited in claims: v0.4.35 (2026-08-26) Pro-plan billing and seat management behind one Cloud address; v0.4.37 (2026-08-31) says Multica installs natively on iPad; v0.5.2 (2026-09-23) and v0.5.3 (2026-09-24) steering of Claude Code, Codex and Grok tasks; v0.5.1 (2026-09-21) issue wakeup rules; v0.6.0 (2026-09-28) wakeups on status change, sub-issue completion and pull request movement, plus add, queue or restart when replying. No dated forward-looking plans in the part read.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

## multica:SHIP-06

Finding: Review happens one issue at a time through the inbox, in_review status and PR cards; batch queues, grouping, risk filters and exact-revision approvals were not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Six-task fixture queue not run; batch features not found.

Source: https://multica.ai/docs/issues
Locator: Parts of an issue; Choose an assignee; Status; Custom statuses; Projects and sub-issues; Switch views
Finding: Re-read 2026-10-09: Seven built-in statuses: backlog, todo, in_progress, in_review, done, blocked, cancelled. Sub-issues with stages and a parent completion notification. No dependency or blocked-by relation documented. Labels and priority described only briefly. When every linked PR merges, the issue moves to the workspace-chosen status, default done.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/github-integration
Locator: Connect GitHub; Link a PR to an issue; View PRs on an issue; When merged PRs move an issue
Finding: Re-read 2026-10-09: GitHub App is read-only and "never pushes commits, comments, or status checks". PRs auto-link by identifier in branch name or title, or with closing keywords in the body. PR card shows state, size, CI status and mergeability. Merging all linked PRs can move the issue to a chosen status. Optional Co-authored-by trailer "multica-agent <github@multica.ai>". No statement on who merges or on importing GitHub Issues, Linear or Jira.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/inbox
Locator: What lands in the inbox; Tune notification types
Finding: Re-read 2026-10-09: Inbox covers assignments, updates on subscribed issues, mentions, reactions, run failures on subscribed issues, results of agent-created issues and autopilot notices. Run-failure notices archive once an issue reaches in_review, started, done or closed; blocked keeps them. Banners follow system and browser settings. Mobile push is not covered on this page.
Retrieved: 2026-10-09

## multica:OWN-01

Finding: Custom Multica License includes extra conditions.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2027-01-07
Plan: Not specified
Surface: Pinned Multica source
Version: commit 8db6cfe

No legal compliance verdict; inspect the full license for a specific use.
The plugin SDK package declares Apache-2.0 separately; hosted parity is unknown.

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/LICENSE
Locator: Part I conditions 1(a) hosted or embedded service, 1(b) branding, 1(c) attribution, 1(d); Part II Apache-2.0 text (333 lines)
Finding: Re-read 2026-10-09: Pinned LICENSE header (8db6cfe): "Multica License", provided under the Apache License 2.0 with additional conditions. Full conditions are summarized on the licensing page re-read today.
Retrieved: 2026-10-09

Source: https://multica.ai/licensing
Locator: Rule of thumb; Common scenarios; Other conditions; Getting a commercial license
Finding: Re-read 2026-10-09: Internal use, including self-hosting for a whole organization, is free. Running for clients, letting outsiders sign in or interact, managed service, or embedding needs a commercial license. Branding and attribution conditions apply. Commercial license requests go through Contact Sales with a stated reply "within three business days".
Retrieved: 2026-10-09

Source: https://multica.ai/about
Locator: Who is behind Multica; links
Finding: Re-read 2026-10-09: Company background, source-available self-hosting, Discord and GitHub community links, and a Contact Sales route. No support email, support SLA or response-time commitment on this page.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/core/types/billing.ts
Locator: Header comment lines 1-12; BillingBatch; BillingTopup
Finding: Re-read 2026-10-09: Pinned billing types (8db6cfe): credit wallet with 1 USD = 1000 credits, credit batches and top-up states. Header states a separate multica-cloud billing service. No public description of what credits purchase.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/packages/plugin-sdk/README.md
Locator: What a surface is; Publishing; What you can do, and what bounds it
Finding: Re-read 2026-10-09: Pinned source plugin-sdk README (8db6cfe): plugin surfaces run in a sandboxed iframe with allow-scripts only; they read issues, post comments authored by the user without @-mention dispatch, and use scoped storage. Catalog gated by plugins_v1.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/issues
Locator: Searches 2026-10-08: Android, App Store, SSO/SAML, dependency/blocked-by, budget/spend limit, cloud runtime
Finding: Re-read 2026-10-09: GitHub searches on 2026-10-09 return 16 open Android results, 12 App Store results and 13 SSO results. Open community requests: #3864 and #3412 (Android), #6996 (mobile push), #7794 and #1218 (blocker dependencies), #6950 (spending limit), #5423 (Azure AAD), #4151 (open-source the hosted runtime). None is a vendor announcement. The repository reports 1,839 open issues and pull requests combined.
Retrieved: 2026-10-09

## multica:OWN-02

Finding: Multica can be self-hosted (Docker Compose, Kubernetes) with Desktop and CLI pointed at it; self-hosted use needs no vendor account, though a daily anonymous telemetry snapshot is on by default. Workspace export and history portability were not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop; Pinned Multica source
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Deployment tested parity, import, cancellation and history portability remain unknown.

Source: https://multica.ai/
Locator: Hero; Works with 20+ AI coding tools; Runtimes; Source available; FAQ
Finding: Re-read 2026-10-09: Names 26 coding tools including ZeroClaw. "Start free trial" button, no price list, "no artificial caps when you self-host", source-available wording. Dashboard mockup shows a "Cloud (Anthropic)" runtime, which the download page describes as not live.
Retrieved: 2026-10-09

Source: https://multica.ai/download
Locator: Choose your platform; Prefer the CLI?; Cloud runtime (waitlist); Current version
Finding: Re-read 2026-10-09: Current version v0.6.1. Installers: macOS Apple Silicon and Intel (.dmg/.zip), Windows x64 and ARM64 (.exe), Linux x64 and ARM64 (.AppImage, .deb, .rpm). CLI install over SSH. Cloud runtime shown as waitlist, "Not live yet". No iOS, Android or minimum OS statement.
Retrieved: 2026-10-09

Source: https://multica.ai/privacy
Locator: Last updated September 24, 2026; AI features; Who we share information with; Where information is stored; How long we keep information
Finding: Re-read 2026-10-09: Last updated 2026-09-24. Operated by Index Labs (Hong Kong) Limited. Hosted on AWS and Vercel. Processors: AWS, Vercel, Stripe, Resend, PostHog, Google sign-in and LLM providers. "Multica does not use your content to train AI models." Cloud chat features send messages to a third-party LLM provider. Privacy requests answered "within 30 days". support@multica.ai is listed for privacy questions and requests. No export of workspace data on exit is stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/desktop-app
Locator: Installation; Built-in daemon; Updating Desktop; Connecting to a self-hosted instance
Finding: Re-read 2026-10-09: Installers: macOS (.dmg), Windows (.exe), Linux (.AppImage, .deb, .rpm). Desktop starts a built-in daemon after sign-in. Auto-update on by default. Self-hosted connection through desktop.json. No minimum OS version stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/cli
Locator: Installation; First connection; Common issue operations; Command reference
Finding: Re-read 2026-10-09: Commands cover issues, projects, labels, agents, skills, autopilots, workspaces, repos, daemon and runtimes. `multica issue cancel-task <task-id> --issue <issue-id>` cancels a run. `multica issue rerun <id>` re-enqueues for the current assignee. Install via curl script, Homebrew or PowerShell. `issue comment update` requires --expected-revision (source, cmd_issue.go).
Retrieved: 2026-10-09

Source: https://multica.ai/docs/self-host-quickstart
Locator: Multica service and computer; make selfhost; Back up Postgres; Kubernetes
Finding: Re-read 2026-10-09: Self-hosting runs Web, API and PostgreSQL through Docker Compose; Kubernetes via Helm is referenced. No email service by default; codes are printed to backend logs until Resend or SMTP is set. Migrations are forward-only. "Self-hosting replaces only the Multica Cloud part."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/environment-variables
Locator: Daemon configuration; Observability and analytics; Server-side LLM
Finding: Re-read 2026-10-09: Telemetry on by default: one anonymous self-host snapshot per UTC day; DO_NOT_TRACK=1 or true disables it; ANALYTICS_DISABLED=true turns off PostHog. MULTICA_DAEMON_MAX_CONCURRENT_TASKS default 20. MULTICA_AGENT_TIMEOUT default 0 (no limit). MULTICA_AGENT_IDLE_WATCHDOG default 2h; tool watchdog same by default. No iteration limit variable listed. Optional server-side LLM for chat titles sends chat content to a configured endpoint.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/SELF_HOSTING.md
Locator: Telemetry paragraph line 19; Kubernetes section
Finding: Re-read 2026-10-09: Pinned SELF_HOSTING.md (8db6cfe): configure RESEND_API_KEY for email codes; without email the code is printed to backend logs; fixed code only when APP_ENV=development.
Retrieved: 2026-10-09

## multica:OWN-03

Finding: Workspace content (issues, comments, run records, attachments) is stored by Multica; code and tool credentials stay on the connected computer except what agents write back; tools send prompts and code to their model providers. The privacy policy states no training use and lists subprocessors; telemetry is documented with opt-outs.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Multica Cloud/web/desktop; Pinned Multica source
Version: v0.6.1 (released 2026-10-01) docs; source commit 8db6cfe

Vendor statements only; no network traffic or first-run consent prompt was observed; chat titles and follow-ups send chat text to a configured LLM endpoint.

Source: https://multica.ai/privacy
Locator: Last updated September 24, 2026; AI features; Who we share information with; Where information is stored; How long we keep information
Finding: Re-read 2026-10-09: Last updated 2026-09-24. Operated by Index Labs (Hong Kong) Limited. Hosted on AWS and Vercel. Processors: AWS, Vercel, Stripe, Resend, PostHog, Google sign-in and LLM providers. "Multica does not use your content to train AI models." Cloud chat features send messages to a third-party LLM provider. Privacy requests answered "within 30 days". support@multica.ai is listed for privacy questions and requests. No export of workspace data on exit is stated.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Re-read 2026-10-09: The daemon is the background process on one computer. A runtime is one computer plus one tool or custom profile. Code and tool credentials stay on the computer; the server stores issues, comments, run records and results. Runtimes offline more than 7 days with no bound agents are cleaned up. Default 20 runs per daemon and 6 per agent. Runtimes are private by default. Custom profiles must fit an existing protocol family; "A custom profile does not add a new communication protocol."
Retrieved: 2026-10-09

Source: https://multica.ai/docs/agents-create
Locator: Starting points; Runtime, model, and thinking level; Configuration after creation; Environment variables and credentials
Finding: Re-read 2026-10-09: Agents start blank or with Build with AI (needs an online runtime). Runtime, model and thinking level. custom_env values are stored in plaintext in the Multica server database; only workspace owners and admins can view or change them. Critical variables such as PATH, HOME and MULTICA_* cannot be overridden.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/environment-variables
Locator: Daemon configuration; Observability and analytics; Server-side LLM
Finding: Re-read 2026-10-09: Telemetry on by default: one anonymous self-host snapshot per UTC day; DO_NOT_TRACK=1 or true disables it; ANALYTICS_DISABLED=true turns off PostHog. MULTICA_DAEMON_MAX_CONCURRENT_TASKS default 20. MULTICA_AGENT_TIMEOUT default 0 (no limit). MULTICA_AGENT_IDLE_WATCHDOG default 2h; tool watchdog same by default. No iteration limit variable listed. Optional server-side LLM for chat titles sends chat content to a configured endpoint.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/install-agent-runtime
Locator: Choose a tool; Sign in on your machine
Finding: Re-read 2026-10-09: Sign in to each tool on its own machine; Multica never receives login tokens. "Supported" means Multica can invoke the CLI; it does not supply the tool account, subscription or model quota. The table lists 25 CLIs. Minimum versions for eight tools; below minimum the runtime is not registered.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/how-multica-works
Locator: The full path of a run; The data and execution boundary
Finding: Re-read 2026-10-09: The server coordinates and records work; connected computers execute it. Code directories and local files stay on the connected computer. The server stores issues, comments, statuses, run records and results. Custom env values are stored on the server. The boundary is the same on Cloud and self-hosted.
Retrieved: 2026-10-09

Source: https://multica.ai/docs/workspaces
Locator: What a workspace contains; Delete a workspace
Finding: Re-read 2026-10-09: A workspace is the isolation boundary for members, issues, projects, agents, skills and run history. Owners and admins manage settings; members view them. Deletion is permanent. No export or portability statement. No SSO mention.
Retrieved: 2026-10-09

Source: https://github.com/multica-ai/multica/blob/8db6cfe19ae6fd5c35ec71bd8fea42a3ef3861ec/SELF_HOSTING.md
Locator: Telemetry paragraph line 19; Kubernetes section
Finding: Re-read 2026-10-09: Pinned SELF_HOSTING.md (8db6cfe): configure RESEND_API_KEY for email codes; without email the code is printed to backend logs; fixed code only when APP_ENV=development.
Retrieved: 2026-10-09

## multica:OWN-04

Finding: No public price list was found: /pricing renders a not-found page and the site offers Get started, Book a demo and Contact sales. The homepage FAQ says use inside your own organization is free, including self-hosting. The changelog mentions subscribing to a Pro plan with billing and seat management, without a published price. Model and compute costs are separate and borne by the user. The public subscription interface confirms monthly/yearly human-seat billing but receives actual prices from the Cloud service.

Verification: sourced
Checked: 2026-11-10
Scheduled recheck: 2026-11-24
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: Live docs reread 2026-11-07 through v0.8.1; source at v0.6.1 commit 8db6cfe

Only the named atomic answers are established by this source pass. Remaining atomic answers retain their documented gaps.

Source: https://multica.ai/
Locator: Hero; Works with 20+ AI coding tools; Runtimes; Source available; FAQ
Finding: Homepage now names 29 coding tools, source-available internal self-hosting, a free-trial button with no numeric trial terms, and connected-computer execution. Self-hosted Qoder Cloud Agents is a separately configured remote runtime.
Retrieved: 2026-11-07

Source: https://multica.ai/download
Locator: Choose your platform; Prefer the CLI?; Cloud runtime (waitlist); Current version
Finding: Downloads show macOS Apple Silicon/Intel, Windows x64/ARM64 and Linux x64/ARM64 installers; current v0.8.1. Vendor-hosted runtime remains not live, waitlist-only. It is separate from connecting Qoder Cloud Agents.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/daemon-runtimes
Locator: Daemon vs runtime; Execution location; Dispatch and online status; Concurrency limits; Private and public runtimes; Custom runtime profiles; Task runtime environment
Finding: Connected-computer daemon claims local-tool runs, stores results on server, and has explicit login autostart. Runtimes offline over seven days with no bound agents are cleaned up. Separate Qoder Cloud Agents bridge uses a PAT and remote provider environment without a local daemon; defaults three concurrent runs and 60-minute timeout, configurable.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/install-agent-runtime
Locator: Choose a tool; Sign in on your machine
Finding: Re-read 2026-10-09: Sign in to each tool on its own machine; Multica never receives login tokens. "Supported" means Multica can invoke the CLI; it does not supply the tool account, subscription or model quota. The table lists 25 CLIs. Minimum versions for eight tools; below minimum the runtime is not registered.
Retrieved: 2026-11-07

Source: https://multica.ai/
Locator: Hero buttons, FAQ "Can I use Multica commercially?", footer and JSON-LD offer
Finding: Offers "Get started", "Book a demo" and a footer "Contact sales" link. FAQ: using Multica inside your own organization is free, including self-hosting; a commercial licence is needed only to offer it to others or embed it in a product. Page metadata lists an offer at price 0 USD. No price list.
Retrieved: 2026-11-10

Source: https://multica.ai/pricing
Locator: Page body
Finding: /pricing renders the site not-found fallback; no price list.
Retrieved: 2026-11-10

Source: https://multica.ai/changelog
Locator: August 26, 2026 entry, Improvements and Bug Fixes
Finding: Says a single Cloud address turns on cloud runtimes, billing and seat management together, and that billing reflects your Pro plan right after you subscribe. No price is given.
Retrieved: 2026-11-10

Source: https://github.com/multica-ai/multica/blob/b7f8efcaae6c5f3273c0038252b5d10c81b41068/packages/core/types/billing.ts
Locator: Workspace subscriptions; WorkspaceSubscriptionPrice, lines 184-280
Finding: Cloud workspace subscriptions have monthly/yearly intervals and purchasable human seats. Displayed prices are unitAmount per licensed seat. The actual price and currency are server-supplied, so example values are not public prices.
Retrieved: 2026-11-13

Source: https://github.com/multica-ai/multica/blob/b7f8efcaae6c5f3273c0038252b5d10c81b41068/apps/docs/content/docs/providers.mdx
Locator: Supported tools; model sources; Amp credentials
Finding: A Multica runtime invokes the coding tool already installed and signed in on the user computer. Tool account access determines available models; Amp explicitly requires AMP_API_KEY in the daemon/agent environment. Model credentials and billing stay with that user/provider account.
Retrieved: 2026-11-13

## multica:OWN-05

Finding: Internal self-hosted software license cost is $0 for one, two or five users. Machines and coding-tool usage remain separate, and no complete workload cost is established.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Internal self-hosted use only; not Cloud trial or commercial hosted licensing
Surface: Multica Cloud/web/desktop
Version: Live docs reread 2026-11-07 through v0.8.1; source at v0.6.1 commit 8db6cfe

No workload, billing or authenticated client test was performed; unestablished details remain unknown.

Source: https://multica.ai/licensing
Locator: Rule of thumb; Common scenarios; Other conditions; Getting a commercial license
Finding: Re-read 2026-10-09: Internal use, including self-hosting for a whole organization, is free. Running for clients, letting outsiders sign in or interact, managed service, or embedding needs a commercial license. Branding and attribution conditions apply. Commercial license requests go through Contact Sales with a stated reply "within three business days".
Retrieved: 2026-11-07

Source: https://multica.ai/docs/install-agent-runtime
Locator: Choose a tool; Sign in on your machine
Finding: Re-read 2026-10-09: Sign in to each tool on its own machine; Multica never receives login tokens. "Supported" means Multica can invoke the CLI; it does not supply the tool account, subscription or model quota. The table lists 25 CLIs. Minimum versions for eight tools; below minimum the runtime is not registered.
Retrieved: 2026-11-07

## multica:OWN-06

Finding: Current release is v0.8.1 dated November 6. Hosted Multica runtime is waitlisted; commercial-license and privacy response windows are distinct from general support guarantees.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: Live docs reread 2026-11-07 through v0.8.1; source at v0.6.1 commit 8db6cfe

Status history and export-on-exit not found; funding and popularity are context only.

Source: https://multica.ai/changelog
Locator: v0.8.1 (November 6, 2026), v0.8.0 (November 3) and historical release entries
Finding: Current changelog lists v0.8.1 dated November 6, 2026, after v0.8.0 November 3. Skills revision history, unread filters and Qoder updates are released, not future promises.
Retrieved: 2026-11-07

Source: https://multica.ai/download
Locator: Choose your platform; Prefer the CLI?; Cloud runtime (waitlist); Current version
Finding: Downloads show macOS Apple Silicon/Intel, Windows x64/ARM64 and Linux x64/ARM64 installers; current v0.8.1. Vendor-hosted runtime remains not live, waitlist-only. It is separate from connecting Qoder Cloud Agents.
Retrieved: 2026-11-07

Source: https://multica.ai/licensing
Locator: Rule of thumb; Common scenarios; Other conditions; Getting a commercial license
Finding: Re-read 2026-10-09: Internal use, including self-hosting for a whole organization, is free. Running for clients, letting outsiders sign in or interact, managed service, or embedding needs a commercial license. Branding and attribution conditions apply. Commercial license requests go through Contact Sales with a stated reply "within three business days".
Retrieved: 2026-11-07

Source: https://multica.ai/privacy
Locator: Last updated October 22, 2026; AI features; retention; choices and rights; contact
Finding: Privacy policy last updated October 22, 2026, read November 7. Operated by Index Labs (Hong Kong) Limited. Cloud hosted on AWS and Vercel; selected Cloud features send recent messages to an LLM provider. Content is not used for AI training. Privacy requests go to support@multica.ai and receive a response within 30 days. Account/workspace content persists while the account/workspace exists, with deletion and backup qualifications. Self-hosted servers send a daily content-free usage snapshot unless DO_NOT_TRACK=1.
Retrieved: 2026-11-07

Source: https://multica.ai/about
Locator: Who is behind Multica; links
Finding: Re-read 2026-10-09: Company background, source-available self-hosting, Discord and GitHub community links, and a Contact Sales route. No support email, support SLA or response-time commitment on this page.
Retrieved: 2026-11-07

Source: https://multica.ai/contact-sales
Locator: Form fields
Finding: Re-read 2026-10-09: Intake form requires a business email (personal providers not accepted), company size band, country and use case. No price, plan or seat terms. No reply-time wording on this page.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/cloud-quickstart
Locator: Steps 1-5; Common problems
Finding: Web/Desktop sign-in uses email code or Google. Desktop connects a computer automatically; other computers install CLI and run setup. At least one installed/authenticated coding tool required; no bundled model quota.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/channels
Locator: Choose a platform; Account binding; Self-hosting
Finding: Re-read 2026-10-09: Feishu/Lark, Slack, DingTalk, WeCom and Telegram. "New connections are currently open only for mainland-China Feishu." Account binding on first message; membership re-checked on every message.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/community-maintained
Locator: Areas table; retirement rule
Finding: Re-read 2026-10-09: DingTalk, WeCom and Telegram areas are community-maintained with "no official support SLA" and no response-time guarantee. Areas that cannot be fixed without platform access may be deprecated rather than left quietly broken.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/issues
Locator: Searches 2026-10-08: Android, App Store, SSO/SAML, dependency/blocked-by, budget/spend limit, cloud runtime
Finding: Newest issue list reread November 7. Requests are not vendor announcements; prior October counts are not refreshed.
Retrieved: 2026-11-07

Source: https://github.com/multica-ai/multica/releases
Locator: Latest release tag
Finding: GitHub latest resolves to v0.8.1 displayed November 6, 2026. Earlier v0.6.1 is historical.
Retrieved: 2026-11-07

## multica:OWN-08

Finding: Self-hosting is free for internal use (infrastructure and tool accounts are the user's), the website offers a Start free trial for Cloud with undisclosed terms, and the hosted cloud runtime is an early-access waitlist.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: Live docs reread 2026-11-07 through v0.8.1; source at v0.6.1 commit 8db6cfe

Card requirement, ongoing free tier and paid evaluation terms are not published.

Source: https://multica.ai/
Locator: Hero; Works with 20+ AI coding tools; Runtimes; Source available; FAQ
Finding: Homepage now names 29 coding tools, source-available internal self-hosting, a free-trial button with no numeric trial terms, and connected-computer execution. Self-hosted Qoder Cloud Agents is a separately configured remote runtime.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/mobile-app
Locator: Intro; Prerequisites; 7-day limit of free signing
Finding: Current mobile guide says source-built iPhone client, no App Store and no Android, seven-day free signing, server-owned workspace data. Native iPad support is separately documented in release notes; no device test.
Retrieved: 2026-11-07

Source: https://multica.ai/download
Locator: Choose your platform; Prefer the CLI?; Cloud runtime (waitlist); Current version
Finding: Downloads show macOS Apple Silicon/Intel, Windows x64/ARM64 and Linux x64/ARM64 installers; current v0.8.1. Vendor-hosted runtime remains not live, waitlist-only. It is separate from connecting Qoder Cloud Agents.
Retrieved: 2026-11-07

Source: https://multica.ai/licensing
Locator: Rule of thumb; Common scenarios; Other conditions; Getting a commercial license
Finding: Re-read 2026-10-09: Internal use, including self-hosting for a whole organization, is free. Running for clients, letting outsiders sign in or interact, managed service, or embedding needs a commercial license. Branding and attribution conditions apply. Commercial license requests go through Contact Sales with a stated reply "within three business days".
Retrieved: 2026-11-07

Source: https://multica.ai/about
Locator: Who is behind Multica; links
Finding: Re-read 2026-10-09: Company background, source-available self-hosting, Discord and GitHub community links, and a Contact Sales route. No support email, support SLA or response-time commitment on this page.
Retrieved: 2026-11-07

Source: https://multica.ai/docs/self-host-quickstart
Locator: Multica service and computer; make selfhost; Back up Postgres; Kubernetes
Finding: Re-read 2026-10-09: Self-hosting runs Web, API and PostgreSQL through Docker Compose; Kubernetes via Helm is referenced. No email service by default; codes are printed to backend logs until Resend or SMTP is set. Migrations are forward-only. "Self-hosting replaces only the Multica Cloud part."
Retrieved: 2026-11-07

Source: https://multica.ai/docs/install-agent-runtime
Locator: Choose a tool; Sign in on your machine
Finding: Re-read 2026-10-09: Sign in to each tool on its own machine; Multica never receives login tokens. "Supported" means Multica can invoke the CLI; it does not supply the tool account, subscription or model quota. The table lists 25 CLIs. Minimum versions for eight tools; below minimum the runtime is not registered.
Retrieved: 2026-11-07

## multica:OWN-09

Finding: A commercial license for external hosted/embedded use is quote-only. No public hosted seat/usage rate was established; internal self-hosting is free under the stated license.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Multica Cloud/web/desktop
Version: Live docs reread 2026-11-07 through v0.8.1; source at v0.6.1 commit 8db6cfe

No workload, billing or authenticated client test was performed; unestablished details remain unknown.

Source: https://multica.ai/licensing
Locator: Rule of thumb; Common scenarios; Other conditions; Getting a commercial license
Finding: Re-read 2026-10-09: Internal use, including self-hosting for a whole organization, is free. Running for clients, letting outsiders sign in or interact, managed service, or embedding needs a commercial license. Branding and attribution conditions apply. Commercial license requests go through Contact Sales with a stated reply "within three business days".
Retrieved: 2026-11-07

Source: https://multica.ai/contact-sales
Locator: Form fields
Finding: Re-read 2026-10-09: Intake form requires a business email (personal providers not accepted), company size band, country and use case. No price, plan or seat terms. No reply-time wording on this page.
Retrieved: 2026-11-07