Back to the Finder

Conductor

Mac app. Agents run on your computer or in the vendor's cloud.

Product websiteconductor.build Compare with others Missing or wrong data?

Conductor product interface, supplied by its publisher
Desktop app
Mac
Browser
Not listed
Phone
Not listed
Command line
Yes
Agents
Claude Code, Codex, Cursor Agent and OpenCode
Your subscription
Works with your own plan
Agents run
On your computer and in the vendor's cloud
Laptop closed
Keeps going
Isolation
Worktree and VM
Parallel agents
Yes
Scheduled runs
Yes
Coordinator agent
Not listed
Built-in issues
Not listed
Pull requests
Yes
Review screen
Yes
Team
Shared sessions and roles
Self-hosting
Not listed
Licence
Proprietary
Price from
Not listed
How it charges
Bring your own key
Paid plans from
Not listed

Receipts

Checked
· oldest answer 2026-10-08
Findings
47 published · 42 cross-checked by a second agent · read them all
Sources
87 public pages on 3 sites: conductor.build, github.com, apps.apple.com
Screenshot
conductor.build, retrieved · first party marketing · shows appearance, not capability
Method
Same rules for all 103 researched ADEs. How we check · Report wrong data

What it does

Where you use it

Conductor documents a macOS app and an iPhone app requiring iOS 26.0+. Release 0.92.0 adds experimental remote control of local Mac workspaces from iPhone. Current installation instructions provide Mac download steps; the announcement no longer states a signup closure.

Source and scope for Where you use it

Agents it runs

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

Source and scope for Agents it runs

Accounts and subscriptions

Subscriptions and API keys are accepted locally and in Cloud; Sign in with ChatGPT was added for Codex; Conductor does not resell model usage.

Source and scope for Accounts and subscriptions

Where agents run

Managed Cloud (Vercel sandbox, us-east-1) keeps agents running after the Mac app closes until sandbox sleep or expiry; local sessions end when the app closes or the Mac shuts down; own and self-hosted cloud machines are not available. Cloud workspace disk is 32 GB; the Cloud Computer disk is 64 GB per one dated changelog entry.

Source and scope for Where agents run

How work is kept apart

Local workspaces use a git worktree and branch each without an OS sandbox; Cloud uses isolated Linux microVM sandboxes.

Source and scope for How work is kept apart

Parallel work

Multiple workspaces, or several chats in one workspace, run in parallel; Home and the sidebar show workspace status, and a matrix-style status indicator was introduced as experimental opt-in (0.85.0) and is shown in cloud organizations per 0.90.0.

Source and scope for Parallel work

Scheduled runs

Cloud routines run saved prompts on schedules (including weekdays), GitHub actions with filters, or webhooks; routines are Pro/Team only.

Source and scope for Scheduled runs

Git and pull requests

Workspace branch-to-PR workflow is documented, with GitHub stacks (0.80.0; local needs the gh stack extension) and an option to open PRs in Graphite (0.86.0).

Source and scope for Git and pull requests

Licence and source

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

Source and scope for Licence and source

Price

The pricing URL redirects to the homepage without a price table. Cloud FAQ describes plan gates and no additional compute fee for now; model charges stay with the provider. Current subscription prices and allowances remain unestablished.

Source and scope for Price

JSON for agentsMarkdown for agents

Report wrong dataSuggest a missing ADE

Typed filters

Every value links to its receipt below. Read plan and surface restrictions before using a filter to decide.

Typed filter values
FilterValue
Desktop OS: macOSYes
Desktop OS: WindowsNo
Desktop OS: LinuxNo
Web clientUnknown
Phone access: Native iOSUnknown
Phone access: Native AndroidUnknown
Phone access: Phone browserUnknown
CLIYes
Coding agents: Claude CodeYes cross-checked
Coding agents: CodexYes cross-checked
Coding agents: Cursor AgentYes cross-checked
Coding agents: Gemini CLINo
Coding agents: OpenCodeYes cross-checked
Coding agents: PiNo
Coding agents: GrokNo
Coding agents: AiderNo
Coding agents: AntigravityNo
Use my subscriptionYes cross-checked
Where agents run: On your computerYes cross-checked
Where agents run: On your own serverNo
Where agents run: In the vendor's cloudYes
Workspace isolation: Git worktreeYes cross-checked
Workspace isolation: ContainerUnknown
Workspace isolation: VM or microVMYes
Parallel runsYes cross-checked
CoordinatorUnknown
Native issue trackingUnknown
Unattended triggersYes
Shared promptingYes
Team rolesYes
SSOUnknown
Pull request workflowYes cross-checked
Review interfaceYes cross-checked
Product license: Apache-2.0No. Proprietary
Product license: MITNo. Proprietary
Product license: GPL-3.0No. Proprietary
Product license: GPL-3.0-or-laterNo. Proprietary
Product license: AGPL-3.0No. Proprietary
Product license: AGPL-3.0-or-laterNo. Proprietary
Product license: Elastic License 2.0No. Proprietary
Product license: Custom termsNo. Proprietary
Product license: ProprietaryYes. Proprietary
Self-hostingUnknown
Telemetry opt-outUnknown
Free accessUnknown
Starting price: FreeUnknown
Starting price: Up to $20/monthUnknown
Starting price: $21–40/monthUnknown
Starting price: $41–100/monthUnknown
Starting price: Over $100/monthUnknown
Starting price: Contact for pricingUnknown
Pricing model: Free, no paid planUnknown
Pricing model: Free planUnknown
Pricing model: SubscriptionUnknown
Pricing model: Per seatUnknown
Pricing model: Usage-basedUnknown
Pricing model: Per machine or computeUnknown
Pricing model: Bring your own keyYes
Pricing model: One-time purchaseUnknown
Paid plans from: No paid planUnknown
Paid plans from: $10 or less/monthUnknown
Paid plans from: $11–20/monthUnknown
Paid plans from: $21–40/monthUnknown
Paid plans from: $41–100/monthUnknown
Paid plans from: Over $100/monthUnknown
Paid plans from: Contact salesUnknown
Lifecycle status: ActiveUnknown
Lifecycle status: SunsettingUnknown
Keeps going with laptop closedYes
Open-source licenseNo. Proprietary
Any desktop appYes
Paid plan up to $20/monthUnknown
Paid plan up to $40/monthUnknown
Any phone clientUnknown

Filter receipts

Desktop OS: macOS: Yes

Desktop OS: macOS

Typed value: Yes.

Conductor documents a macOS app and an iPhone app requiring iOS 26.0+. Release 0.92.0 adds experimental remote control of local Mac workspaces from iPhone. Current installation instructions provide Mac download steps; the announcement no longer states a signup closure.

  • Windows/Linux are explicitly unavailable; Android remains undocumented.
  • Remote Control requires the Mac and experimental setting. No app or authenticated Cloud test was run.
  1. Install Conductor

    Finding: The install guide provides Mac download/setup steps, explicitly excludes Windows/Linux, and requires GitHub and an agent provider. It does not state an existing-account-only restriction.

Desktop OS: Windows: No

Desktop OS: Windows

Typed value: No.

Conductor documents a macOS app and an iPhone app requiring iOS 26.0+. Release 0.92.0 adds experimental remote control of local Mac workspaces from iPhone. Current installation instructions provide Mac download steps; the announcement no longer states a signup closure.

  • Windows/Linux are explicitly unavailable; Android remains undocumented.
  • Remote Control requires the Mac and experimental setting. No app or authenticated Cloud test was run.
  1. Install Conductor

    Finding: The install guide provides Mac download/setup steps, explicitly excludes Windows/Linux, and requires GitHub and an agent provider. It does not state an existing-account-only restriction.

Desktop OS: Linux: No

Desktop OS: Linux

Typed value: No.

Conductor documents a macOS app and an iPhone app requiring iOS 26.0+. Release 0.92.0 adds experimental remote control of local Mac workspaces from iPhone. Current installation instructions provide Mac download steps; the announcement no longer states a signup closure.

  • Windows/Linux are explicitly unavailable; Android remains undocumented.
  • Remote Control requires the Mac and experimental setting. No app or authenticated Cloud test was run.
  1. Install Conductor

    Finding: The install guide provides Mac download/setup steps, explicitly excludes Windows/Linux, and requires GitHub and an agent provider. It does not state an existing-account-only restriction.

CLI: Yes

CLI

Typed value: Yes.

Beta HTTP API (/v0, OpenAPI), hosted Streamable HTTP MCP server and routine webhooks; Conductor CLI exists in cloud workspaces.

  • Cli: The Conductor CLI is preinstalled in cloud workspaces (0.80.0) and can manage sidebar sections in cloud organizations (0.87.0); the API doc shows a routine example with the `conductor` CLI using CONDUCTOR_API_KEY. No install guide or command reference located.
  • Api: HTTP /v0.
  • Mcp server: Hosted Streamable HTTP.
  • Webhooks: Routines expose a webhook URL that starts a saved prompt in a fresh workspace on JSON POST; webhooks are routine-only and the URL is the credential.
  • Operations: API/MCP: list projects; create, get, rename, archive, unarchive and sleep workspaces; create and rename sessions; send and cancel prompts; read transcripts; read-only SQL; routines (API).
  • Authentication: Bearer API key or organization-scoped MCP OAuth.
  • Stability: Beta interfaces may change.
  • API and MCP are labelled beta and may change.
  • No SDK located. Public issue #32 (opened 2026-09-19, open, no comments) reports an API task message returning HTTP 403 "Blocked". The API doc attributes some 403s to a proxy that rejects default client signatures and advises sending a User-Agent. Not reproduced here.
  • The API Quick start requires a Conductor Pro account.
  1. Conductor API

    Finding: Beta HTTP API under https://api.conductor.build/v0 with an OpenAPI document; bearer API key; endpoints for projects, workspaces (create, get, rename, archive, unarchive, sleep), sessions, prompts, transcripts, cancel ("drop queued messages"), SQL and routines. Routines created through the API run a saved prompt in a fresh workspace on each webhook JSON POST. Cookbook: "messages sent while it works are steered into the running turn"; Tips: a queued prompt reports idle until its turn starts. Some 403s come from a proxy that rejects default client signatures (send a User-Agent). Shows a routine example with the `conductor` CLI. No SDK is documented on this page. Quick start (for AI) requires a Conductor Pro account to generate an API key. Page undated. Endpoint table: GET /v0/sessions/{id}/status "Whether the agent is idle, working, or errored"; POST /v0/sessions/{id}/cancel "Stop the agent's current turn and drop queued messages". Tips: "Always pass an explicit model" (workspace creation, POST /v0/workspaces) and "An optional effort field sets the thinking level for claude and codex sessions". No sidebar-section endpoint listed.

  2. Conductor 0.80.0: Stacks

    Finding: GitHub stacks (cloud; local needs gh stack extension); cloud workspaces include the Conductor CLI and Chrome; cloud environment variables moved to Organization → Environment and "editing them no longer requires rebuilding your Cloud Computer"; GH_HOST respected for GitHub Enterprise.

Coding agents: Claude Code: Yes cross-checked

Coding agents: Claude Code

Typed value: Yes.

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

  • Harnesses: Claude Code, Codex, Cursor and OpenCode.
  • Model providers: Claude Code with Anthropic, OpenRouter, Vercel AI Gateway, AWS Bedrock or GLM via environment variables; Codex via OpenAI/ChatGPT or a gateway; OpenCode via provider API keys (OpenRouter, Baseten, Cerebras, AI Gateway); Cursor via Cursor API key.
  • Session integration: Each chat tab runs through one harness and model; Claude Code, Codex and OpenCode are bundled, and Cursor sessions use Cursor Agent through the Cursor API.
  • Mixed harness operation: Documented, not observed: several agents can run in separate chat tabs of one workspace ("Run multiple agents in one workspace when the work belongs on the same branch") or in separate workspaces, with one harness and model per chat tab. Not tested.
  • Cursor runs through the Cursor API rather than a local executable.
  • Provider coverage per harness and in Cloud was not tested; Vertex AI is named only as a Claude provider setting (settings reference) with no setup guide located, and custom harness adapters are not documented.
  • Mixed-harness operation is documented but was not observed.
  1. Conductor

    Finding: Vendor homepage presents a team of coding agents running in the cloud, isolated cloud workspaces, and shared workspaces where teammates can follow work and prompt together in real time. It names Claude Code, Codex, Cursor and OpenCode. The Conduct from anywhere section links macOS, the iOS App Store and the API; the latest release link is 0.90.0 (Conductor for iOS, 2026-10-02).

  2. Parallel agents

    Finding: Independent branches or multiple agents sharing one checkout; shared-file collision risk is explicit.

  3. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

  4. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

  5. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  6. Settings reference

    Finding: Settings include tool_approvals_enabled, Codex tool_approval_policy (default or always_approve), default plan mode, review/PR/conflict prompts, git.archive_on_merge, enterprise_data_privacy.

Coding agents: Codex: Yes cross-checked

Coding agents: Codex

Typed value: Yes.

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

  • Harnesses: Claude Code, Codex, Cursor and OpenCode.
  • Model providers: Claude Code with Anthropic, OpenRouter, Vercel AI Gateway, AWS Bedrock or GLM via environment variables; Codex via OpenAI/ChatGPT or a gateway; OpenCode via provider API keys (OpenRouter, Baseten, Cerebras, AI Gateway); Cursor via Cursor API key.
  • Session integration: Each chat tab runs through one harness and model; Claude Code, Codex and OpenCode are bundled, and Cursor sessions use Cursor Agent through the Cursor API.
  • Mixed harness operation: Documented, not observed: several agents can run in separate chat tabs of one workspace ("Run multiple agents in one workspace when the work belongs on the same branch") or in separate workspaces, with one harness and model per chat tab. Not tested.
  • Cursor runs through the Cursor API rather than a local executable.
  • Provider coverage per harness and in Cloud was not tested; Vertex AI is named only as a Claude provider setting (settings reference) with no setup guide located, and custom harness adapters are not documented.
  • Mixed-harness operation is documented but was not observed.
  1. Conductor

    Finding: Vendor homepage presents a team of coding agents running in the cloud, isolated cloud workspaces, and shared workspaces where teammates can follow work and prompt together in real time. It names Claude Code, Codex, Cursor and OpenCode. The Conduct from anywhere section links macOS, the iOS App Store and the API; the latest release link is 0.90.0 (Conductor for iOS, 2026-10-02).

  2. Parallel agents

    Finding: Independent branches or multiple agents sharing one checkout; shared-file collision risk is explicit.

  3. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

  4. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

  5. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  6. Settings reference

    Finding: Settings include tool_approvals_enabled, Codex tool_approval_policy (default or always_approve), default plan mode, review/PR/conflict prompts, git.archive_on_merge, enterprise_data_privacy.

Coding agents: Cursor Agent: Yes cross-checked

Coding agents: Cursor Agent

Typed value: Yes.

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

  • Harnesses: Claude Code, Codex, Cursor and OpenCode.
  • Model providers: Claude Code with Anthropic, OpenRouter, Vercel AI Gateway, AWS Bedrock or GLM via environment variables; Codex via OpenAI/ChatGPT or a gateway; OpenCode via provider API keys (OpenRouter, Baseten, Cerebras, AI Gateway); Cursor via Cursor API key.
  • Session integration: Each chat tab runs through one harness and model; Claude Code, Codex and OpenCode are bundled, and Cursor sessions use Cursor Agent through the Cursor API.
  • Mixed harness operation: Documented, not observed: several agents can run in separate chat tabs of one workspace ("Run multiple agents in one workspace when the work belongs on the same branch") or in separate workspaces, with one harness and model per chat tab. Not tested.
  • Cursor runs through the Cursor API rather than a local executable.
  • Provider coverage per harness and in Cloud was not tested; Vertex AI is named only as a Claude provider setting (settings reference) with no setup guide located, and custom harness adapters are not documented.
  • Mixed-harness operation is documented but was not observed.
  1. Conductor

    Finding: Vendor homepage presents a team of coding agents running in the cloud, isolated cloud workspaces, and shared workspaces where teammates can follow work and prompt together in real time. It names Claude Code, Codex, Cursor and OpenCode. The Conduct from anywhere section links macOS, the iOS App Store and the API; the latest release link is 0.90.0 (Conductor for iOS, 2026-10-02).

  2. Parallel agents

    Finding: Independent branches or multiple agents sharing one checkout; shared-file collision risk is explicit.

  3. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

  4. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

  5. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  6. Settings reference

    Finding: Settings include tool_approvals_enabled, Codex tool_approval_policy (default or always_approve), default plan mode, review/PR/conflict prompts, git.archive_on_merge, enterprise_data_privacy.

Coding agents: Gemini CLI: No

Coding agents: Gemini CLI

Typed value: No.

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

  • Harnesses: Claude Code, Codex, Cursor and OpenCode.
  • Session integration: Each chat tab runs through one harness and model; Claude Code, Codex and OpenCode are bundled, and Cursor sessions use Cursor Agent through the Cursor API.
  • Mixed harness operation: Documented, not observed: several agents can run in separate chat tabs of one workspace ("Run multiple agents in one workspace when the work belongs on the same branch") or in separate workspaces, with one harness and model per chat tab. Not tested.
  • Cursor runs through the Cursor API rather than a local executable.
  • Provider coverage per harness and in Cloud was not tested; Vertex AI is named only as a Claude provider setting (settings reference) with no setup guide located, and custom harness adapters are not documented.
  • Mixed-harness operation is documented but was not observed.
  1. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

Coding agents: OpenCode: Yes cross-checked

Coding agents: OpenCode

Typed value: Yes.

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

  • Harnesses: Claude Code, Codex, Cursor and OpenCode.
  • Model providers: Claude Code with Anthropic, OpenRouter, Vercel AI Gateway, AWS Bedrock or GLM via environment variables; Codex via OpenAI/ChatGPT or a gateway; OpenCode via provider API keys (OpenRouter, Baseten, Cerebras, AI Gateway); Cursor via Cursor API key.
  • Session integration: Each chat tab runs through one harness and model; Claude Code, Codex and OpenCode are bundled, and Cursor sessions use Cursor Agent through the Cursor API.
  • Mixed harness operation: Documented, not observed: several agents can run in separate chat tabs of one workspace ("Run multiple agents in one workspace when the work belongs on the same branch") or in separate workspaces, with one harness and model per chat tab. Not tested.
  • Cursor runs through the Cursor API rather than a local executable.
  • Provider coverage per harness and in Cloud was not tested; Vertex AI is named only as a Claude provider setting (settings reference) with no setup guide located, and custom harness adapters are not documented.
  • Mixed-harness operation is documented but was not observed.
  1. Conductor

    Finding: Vendor homepage presents a team of coding agents running in the cloud, isolated cloud workspaces, and shared workspaces where teammates can follow work and prompt together in real time. It names Claude Code, Codex, Cursor and OpenCode. The Conduct from anywhere section links macOS, the iOS App Store and the API; the latest release link is 0.90.0 (Conductor for iOS, 2026-10-02).

  2. Parallel agents

    Finding: Independent branches or multiple agents sharing one checkout; shared-file collision risk is explicit.

  3. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

  4. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

  5. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  6. Settings reference

    Finding: Settings include tool_approvals_enabled, Codex tool_approval_policy (default or always_approve), default plan mode, review/PR/conflict prompts, git.archive_on_merge, enterprise_data_privacy.

Coding agents: Pi: No

Coding agents: Pi

Typed value: No.

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

  • Harnesses: Claude Code, Codex, Cursor and OpenCode.
  • Session integration: Each chat tab runs through one harness and model; Claude Code, Codex and OpenCode are bundled, and Cursor sessions use Cursor Agent through the Cursor API.
  • Mixed harness operation: Documented, not observed: several agents can run in separate chat tabs of one workspace ("Run multiple agents in one workspace when the work belongs on the same branch") or in separate workspaces, with one harness and model per chat tab. Not tested.
  • Cursor runs through the Cursor API rather than a local executable.
  • Provider coverage per harness and in Cloud was not tested; Vertex AI is named only as a Claude provider setting (settings reference) with no setup guide located, and custom harness adapters are not documented.
  • Mixed-harness operation is documented but was not observed.
  1. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

Coding agents: Grok: No

Coding agents: Grok

Typed value: No.

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

  • Harnesses: Claude Code, Codex, Cursor and OpenCode.
  • Session integration: Each chat tab runs through one harness and model; Claude Code, Codex and OpenCode are bundled, and Cursor sessions use Cursor Agent through the Cursor API.
  • Mixed harness operation: Documented, not observed: several agents can run in separate chat tabs of one workspace ("Run multiple agents in one workspace when the work belongs on the same branch") or in separate workspaces, with one harness and model per chat tab. Not tested.
  • Cursor runs through the Cursor API rather than a local executable.
  • Provider coverage per harness and in Cloud was not tested; Vertex AI is named only as a Claude provider setting (settings reference) with no setup guide located, and custom harness adapters are not documented.
  • Mixed-harness operation is documented but was not observed.
  1. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

Coding agents: Aider: No

Coding agents: Aider

Typed value: No.

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

  • Harnesses: Claude Code, Codex, Cursor and OpenCode.
  • Session integration: Each chat tab runs through one harness and model; Claude Code, Codex and OpenCode are bundled, and Cursor sessions use Cursor Agent through the Cursor API.
  • Mixed harness operation: Documented, not observed: several agents can run in separate chat tabs of one workspace ("Run multiple agents in one workspace when the work belongs on the same branch") or in separate workspaces, with one harness and model per chat tab. Not tested.
  • Cursor runs through the Cursor API rather than a local executable.
  • Provider coverage per harness and in Cloud was not tested; Vertex AI is named only as a Claude provider setting (settings reference) with no setup guide located, and custom harness adapters are not documented.
  • Mixed-harness operation is documented but was not observed.
  1. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

Coding agents: Antigravity: No

Coding agents: Antigravity

Typed value: No.

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

  • Harnesses: Claude Code, Codex, Cursor and OpenCode.
  • Session integration: Each chat tab runs through one harness and model; Claude Code, Codex and OpenCode are bundled, and Cursor sessions use Cursor Agent through the Cursor API.
  • Mixed harness operation: Documented, not observed: several agents can run in separate chat tabs of one workspace ("Run multiple agents in one workspace when the work belongs on the same branch") or in separate workspaces, with one harness and model per chat tab. Not tested.
  • Cursor runs through the Cursor API rather than a local executable.
  • Provider coverage per harness and in Cloud was not tested; Vertex AI is named only as a Claude provider setting (settings reference) with no setup guide located, and custom harness adapters are not documented.
  • Mixed-harness operation is documented but was not observed.
  1. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

Use my subscription: Yes cross-checked

Use my subscription

Typed value: Yes.

Subscriptions and API keys are accepted locally and in Cloud; Sign in with ChatGPT was added for Codex; Conductor does not resell model usage.

  • Agent subscriptions: Claude subscription or CLI login, Codex sign-in including Sign in with ChatGPT (0.89.0), Cursor subscription for cloud workspaces (0.85.0); Cloud accepts subscriptions.
  • Api accounts: API keys for Anthropic, OpenAI/Codex, Cursor and OpenCode providers; Cloud accepts keys and org admins can share keys.
  • Account switching: Settings switch between subscription and API-key mode; cloud agent auth can be swapped without restart (0.84.1) and chosen per organization (0.85.0).
  • Vendor credits and provider restrictions remain unestablished; the OpenCode local-model route is documented.
  • Credential flows were not exercised.
  1. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  2. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  3. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  4. Conductor 0.89.0: Sign in with ChatGPT

    Finding: Sign in with ChatGPT for Codex; "Steer Cursor chats while they're running, in local and cloud workspaces."; PR view top card; merge blocked if the PR changed on GitHub; Codex tasks no longer stop on temporary rate-limit errors; custom OAuth MCP servers can use the Conductor callback URL. PR view: "open the Review row to add unresolved comments to chat, or the Checks row to re-run or fix failed checks."

Where agents run: On your computer: Yes cross-checked

Where agents run: On your computer

Typed value: Yes.

Managed Cloud (Vercel sandbox, us-east-1) keeps agents running after the Mac app closes until sandbox sleep or expiry; local sessions end when the app closes or the Mac shuts down; own and self-hosted cloud machines are not available. Cloud workspace disk is 32 GB; the Cloud Computer disk is 64 GB per one dated changelog entry.

  • Payer: Users bring provider subscriptions or API keys. The Cloud FAQ says there is currently no additional compute fee; current subscription prices are not established.
  • Sandbox sleeps after 20 minutes without agent or terminal activity and stops at 23h50m.
  • Disk sizes are scoped: cloud workspaces 32 GB (undated docs); Cloud Computer 64 GB from one dated changelog entry that the docs do not repeat. No hands-on check of either disk.
  • Behavior across a disconnect was not tested.
  • Continuation from another device is documented but was not tested.
  1. Conductor Cloud FAQ

    Finding: Cloud access is described for Pro, Teams and Enterprise; Team signup is invite-only. The FAQ states no additional Cloud compute fee currently and plans usage-based compute pricing. Users bring subscriptions or keys. It also describes managed Vercel sandboxes and a planned self-hosted offering, without release dates. The mobile coming-soon wording is stale against the released iOS app.

  2. What is Conductor Cloud?

    Finding: The Mac app supports local and Cloud workspaces. Local processes stop when the Mac closes; Cloud agents continue until sandbox sleep. The page still documents shared Cloud workspaces and setup.

Where agents run: On your own server: No

Where agents run: On your own server

Typed value: No.

Managed Cloud (Vercel sandbox, us-east-1) keeps agents running after the Mac app closes until sandbox sleep or expiry; local sessions end when the app closes or the Mac shuts down; own and self-hosted cloud machines are not available. Cloud workspace disk is 32 GB; the Cloud Computer disk is 64 GB per one dated changelog entry.

  • Payer: Users bring provider subscriptions or API keys. The Cloud FAQ says there is currently no additional compute fee; current subscription prices are not established.
  • Sandbox sleeps after 20 minutes without agent or terminal activity and stops at 23h50m.
  • Disk sizes are scoped: cloud workspaces 32 GB (undated docs); Cloud Computer 64 GB from one dated changelog entry that the docs do not repeat. No hands-on check of either disk.
  • Behavior across a disconnect was not tested.
  • Continuation from another device is documented but was not tested.
  1. Conductor Cloud FAQ

    Finding: Cloud access is described for Pro, Teams and Enterprise; Team signup is invite-only. The FAQ states no additional Cloud compute fee currently and plans usage-based compute pricing. Users bring subscriptions or keys. It also describes managed Vercel sandboxes and a planned self-hosted offering, without release dates. The mobile coming-soon wording is stale against the released iOS app.

Where agents run: In the vendor's cloud: Yes

Where agents run: In the vendor's cloud

Typed value: Yes.

Managed Cloud (Vercel sandbox, us-east-1) keeps agents running after the Mac app closes until sandbox sleep or expiry; local sessions end when the app closes or the Mac shuts down; own and self-hosted cloud machines are not available. Cloud workspace disk is 32 GB; the Cloud Computer disk is 64 GB per one dated changelog entry.

  • Payer: Users bring provider subscriptions or API keys. The Cloud FAQ says there is currently no additional compute fee; current subscription prices are not established.
  • Sandbox sleeps after 20 minutes without agent or terminal activity and stops at 23h50m.
  • Disk sizes are scoped: cloud workspaces 32 GB (undated docs); Cloud Computer 64 GB from one dated changelog entry that the docs do not repeat. No hands-on check of either disk.
  • Behavior across a disconnect was not tested.
  • Continuation from another device is documented but was not tested.
  1. Conductor Cloud FAQ

    Finding: Cloud access is described for Pro, Teams and Enterprise; Team signup is invite-only. The FAQ states no additional Cloud compute fee currently and plans usage-based compute pricing. Users bring subscriptions or keys. It also describes managed Vercel sandboxes and a planned self-hosted offering, without release dates. The mobile coming-soon wording is stale against the released iOS app.

  2. What is Conductor Cloud?

    Finding: The Mac app supports local and Cloud workspaces. Local processes stop when the Mac closes; Cloud agents continue until sandbox sleep. The page still documents shared Cloud workspaces and setup.

Workspace isolation: Git worktree: Yes cross-checked

Workspace isolation: Git worktree

Typed value: Yes.

Local workspaces use a git worktree and branch each without an OS sandbox; Cloud uses isolated Linux microVM sandboxes.

  • Branch: Each workspace has its own branch; a branch can be checked out in only one workspace at a time.
  • Process: Each workspace has its own running environment and CONDUCTOR_PORT locally.
  • Code separation: Each workspace has separate files, branch and app processes.
  • Network isolation and container use not documented.
  • Isolation is described as development isolation, not a security boundary.
  • Security page says enabling permissions lets cloud agents act on the local Mac.
  1. Isolated workspaces

    Finding: Each workspace is a git worktree on its own branch; a branch can be checked out in one workspace at a time; states workspace isolation is development isolation, not a security boundary, and agents run on the Mac with user permissions.

Workspace isolation: VM or microVM: Yes

Workspace isolation: VM or microVM

Typed value: Yes.

Local workspaces use a git worktree and branch each without an OS sandbox; Cloud uses isolated Linux microVM sandboxes.

  • Vm: Cloud workspaces run in isolated microVM Vercel sandboxes.
  • Secrets: Cloud secrets are hidden in settings but exposed as environment variables to agents inside workspaces.
  • Network isolation and container use not documented.
  • Isolation is described as development isolation, not a security boundary.
  • Security page says enabling permissions lets cloud agents act on the local Mac.
  1. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  2. Cloud Computer

    Finding: Each cloud workspace runs in an isolated microVM with 8 vCPU, 16 GB RAM and 32 GB ephemeral NVMe. The Cloud Computer is the shared development environment for a Conductor organization; the page gives no Cloud Computer disk size; up to 50 GitHub repositories; successful builds seed new workspaces; saved secrets are hidden but exposed as env vars inside workspaces; an Admin workspace agent can edit install/setup scripts and run builds.

Parallel runs: Yes cross-checked

Parallel runs

Typed value: Yes.

Multiple workspaces, or several chats in one workspace, run in parallel; Home and the sidebar show workspace status, and a matrix-style status indicator was introduced as experimental opt-in (0.85.0) and is shown in cloud organizations per 0.90.0.

  • Concurrent sessions: Same-workspace sessions documented.
  • Concurrent tasks: Independent workspaces documented.
  • Concurrency quotas and large-fleet behavior were not tested; same-workspace agents can edit the same files.
  1. Parallel agents

    Finding: Independent branches or multiple agents sharing one checkout; shared-file collision risk is explicit.

Unattended triggers: Yes

Unattended triggers

Typed value: Yes.

Cloud routines run saved prompts on schedules (including weekdays), GitHub actions with filters, or webhooks; routines are Pro/Team only.

  • Schedules: Scheduled routines including a Monday-Friday option (0.87.0).
  • Events: GitHub action triggers with label, author and base-branch filters (0.87.0) and webhook triggers via the API.
  • Recurring tasks: Routines run a saved prompt in a fresh workspace each time.
  • Scheduler location: Routines run in the cloud.
  • Client independence: Routines run in the cloud rather than on the Mac (0.85.0), and Cloud workspaces keep working after the app closes.
  • Overlap, retry and completion behavior are not documented.
  • Routines require Pro/Team and run only in Cloud.
  1. Conductor API

    Finding: Beta HTTP API under https://api.conductor.build/v0 with an OpenAPI document; bearer API key; endpoints for projects, workspaces (create, get, rename, archive, unarchive, sleep), sessions, prompts, transcripts, cancel ("drop queued messages"), SQL and routines. Routines created through the API run a saved prompt in a fresh workspace on each webhook JSON POST. Cookbook: "messages sent while it works are steered into the running turn"; Tips: a queued prompt reports idle until its turn starts. Some 403s come from a proxy that rejects default client signatures (send a User-Agent). Shows a routine example with the `conductor` CLI. No SDK is documented on this page. Quick start (for AI) requires a Conductor Pro account to generate an API key. Page undated. Endpoint table: GET /v0/sessions/{id}/status "Whether the agent is idle, working, or errored"; POST /v0/sessions/{id}/cancel "Stop the agent's current turn and drop queued messages". Tips: "Always pass an explicit model" (workspace creation, POST /v0/workspaces) and "An optional effort field sets the thinking level for claude and codex sessions". No sidebar-section endpoint listed.

  2. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  3. Conductor 0.87.0: Search across files, OpenCode 2.0

    Finding: Routine GitHub trigger filters (label, author, base branch) and weekday schedules; OpenCode 2.0; "Create and delete sidebar sections from the CLI, API, and MCP in cloud organizations"; with the experimental status matrix enabled, indicators "show distinct states for routine runs, attention, errors, running pull request checks, and merged pull requests"; video thumbnails in chat; security fixes; setup scripts from shared cloud workspaces no longer auto-run on collaborators' Macs.

Shared prompting: Yes

Shared prompting

Typed value: Yes.

Shared Cloud workspaces support live chats, presence and typing indicators; private workspaces exist; Pro collaboration covers up to five Pro users.

  • Coprompting: Rests on the undated Cloud FAQ ("collaborate live by sharing workspaces and prompting agents together in real time") and changelog 0.78.0 (2026-07-30, "prompt agents together in real time"). The collaboration doc covers only typing indicators in a shared chat. Input ordering was not tested.
  • Input ordering and conflicts were not tested.
  1. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  2. Conductor 0.78.0: Introducing Conductor Cloud

    Finding: Cloud workspaces on isolated microVMs that keep running when the laptop closes; start by signing up for Pro and creating an organization; shareable links, presence and teammates prompting agents together in real time; API to manage cloud workspaces. Said the mobile app was "coming soon" on 2026-07-30. "To get started, sign up for Pro and create an organization." Billing "opens the Stripe billing portal". No price stated.

Team roles: Yes

Team roles

Typed value: Yes.

Cloud organizations have admins and members. Organization admins can configure shared agent API keys, while members provide personal credentials. Current SSO protocol and enterprise provisioning details are not reestablished by this pass.

  • Organization roles: Organization admins have the documented additional permission to add organization-wide agent keys; members use personal credentials.
  • Membership: The onboarding flow creates or selects an organization and optionally invites teammates; joining members can enter already configured workspaces with their own credentials.
  • Plan scope: Cloud access is a prerequisite. Admin-versus-member credential permissions are explicit in onboarding; a current full paid-plan matrix is unestablished.
  • First-party documentation review only; no multi-account permission or enterprise-login scenario was run. Retained task-role findings keep their older cited scope.
  1. Conductor Cloud organization onboarding

    Finding: Members joining an already configured organization add their agent credentials as needed. Organization admins can additionally configure shared agent API keys for the organization; those keys override a member personal key for the same agent. Organization creation includes optional teammate invitations.

Pull request workflow: Yes cross-checked

Pull request workflow

Typed value: Yes.

Workspace branch-to-PR workflow is documented, with GitHub stacks (0.80.0; local needs the gh stack extension) and an option to open PRs in Graphite (0.86.0).

  • Pull requests: Guided Create PR from the diff viewer; 0.86.0: "Open pull requests in Graphite. Choose it under Open pull requests in Settings → Git".
  • Comments: Review comments return to agent.
  • Ci response: Diff viewer: recommended actions "create a pull request, respond to feedback, fix checks, and merge"; 0.89.0 PR view: "the Checks row to re-run or fix failed checks". What the fix action sends to the agent is not described; not tested.
  • Rebase and commit handling not documented.
  • The fix-checks action is documented without describing what it sends to the agent; not tested.
  1. Diff viewer

    Finding: Unified/commit-filtered diffs, line feedback and GitHub review comments. "Conductor recommends actions as a workspace moves toward merge. Use those actions to create a pull request, respond to feedback, fix checks, and merge". Page undated.

  2. Conductor 0.86.0: Share your loadouts

    Finding: "Removed Codex goals and message queuing. Messages sent while an agent is working now steer it instead of waiting in a queue" (no surface named); answer Codex questions while it works (no surface named); context usage popover shows Claude account limits and reset times, matching Codex; cloud run scripts; private workspaces marked with a lock; fixes for OpenCode rejecting Always allow approvals and for "cloud plan approvals appearing successful when the workspace rejected the response". Improvements also: "Open pull requests in Graphite"; "Prompt links fill the new-workspace composer in your active organization." The Always allow fix is in the Fixes list.

Review interface: Yes cross-checked

Review interface

Typed value: Yes.

Cloud workspaces show diffs; plans, screenshots and attachments in .context sync from cloud workspaces to the Mac (0.85.0); chat videos show a thumbnail that opens the media viewer (0.87.0).

  • Diffs: Diff viewer with unified and per-commit views; files can be edited in diffs (0.84.0).
  • Recordings, test outputs not documented in detail.
  1. Diff viewer

    Finding: Unified/commit-filtered diffs, line feedback and GitHub review comments. "Conductor recommends actions as a workspace moves toward merge. Use those actions to create a pull request, respond to feedback, fix checks, and merge". Page undated.

  2. Conductor 0.84.0: Edit files directly in diffs

    Finding: Edit files in diffs; effort change mid-chat for some Claude models; "Codex chats now retry automatically when OpenAI rate-limits a request" (no surface named); "Port forwarding is now off by default for each workspace"; PR links can open in GitHub or Linear (Settings → Git); 0.84.1 swaps cloud agent auth without restart; remote desktop control in cloud (experimental).

Product license: Apache-2.0: No. Proprietary

Product license: Apache-2.0

Typed value: No. Proprietary.

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Client license: Proprietary vendor software/service terms; source-visible integrations do not license the main application.
  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

Product license: MIT: No. Proprietary

Product license: MIT

Typed value: No. Proprietary.

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Client license: Proprietary vendor software/service terms; source-visible integrations do not license the main application.
  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

Product license: GPL-3.0: No. Proprietary

Product license: GPL-3.0

Typed value: No. Proprietary.

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Client license: Proprietary vendor software/service terms; source-visible integrations do not license the main application.
  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

Product license: GPL-3.0-or-later: No. Proprietary

Product license: GPL-3.0-or-later

Typed value: No. Proprietary.

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Client license: Proprietary vendor software/service terms; source-visible integrations do not license the main application.
  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

Product license: AGPL-3.0: No. Proprietary

Product license: AGPL-3.0

Typed value: No. Proprietary.

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Client license: Proprietary vendor software/service terms; source-visible integrations do not license the main application.
  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

Product license: AGPL-3.0-or-later: No. Proprietary

Product license: AGPL-3.0-or-later

Typed value: No. Proprietary.

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Client license: Proprietary vendor software/service terms; source-visible integrations do not license the main application.
  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

Product license: Elastic License 2.0: No. Proprietary

Product license: Elastic License 2.0

Typed value: No. Proprietary.

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Client license: Proprietary vendor software/service terms; source-visible integrations do not license the main application.
  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

Product license: Custom terms: No. Proprietary

Product license: Custom terms

Typed value: No. Proprietary.

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Client license: Proprietary vendor software/service terms; source-visible integrations do not license the main application.
  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

Product license: Proprietary: Yes. Proprietary

Product license: Proprietary

Typed value: Yes. Proprietary.

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Client license: Proprietary vendor software/service terms; source-visible integrations do not license the main application.
  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

Pricing model: Bring your own key: Yes

Pricing model: Bring your own key

Typed value: Yes.

The pricing URL redirects to the homepage without a price table. Cloud FAQ describes plan gates and no additional compute fee for now; model charges stay with the provider. Current subscription prices and allowances remain unestablished.

  • Model cost: The selected provider account or subscription bills model usage; Conductor does not resell it.
  • Public download and plan documentation does not establish signup completion, current charges or upgrade eligibility. No authenticated billing test was performed.
  1. Configure model providers

    Finding: Model usage is billed through the selected provider account, subscription, usage pool or key. Conductor does not bill or resell it.

Keeps going with laptop closed: Yes

Keeps going with laptop closed

Typed value: Yes.

Managed Cloud (Vercel sandbox, us-east-1) keeps agents running after the Mac app closes until sandbox sleep or expiry; local sessions end when the app closes or the Mac shuts down; own and self-hosted cloud machines are not available. Cloud workspace disk is 32 GB; the Cloud Computer disk is 64 GB per one dated changelog entry.

  • Sandbox sleeps after 20 minutes without agent or terminal activity and stops at 23h50m.
  • Disk sizes are scoped: cloud workspaces 32 GB (undated docs); Cloud Computer 64 GB from one dated changelog entry that the docs do not repeat. No hands-on check of either disk.
  • Behavior across a disconnect was not tested.
  • Continuation from another device is documented but was not tested.
  1. What is Conductor Cloud?

    Finding: The Mac app supports local and Cloud workspaces. Local processes stop when the Mac closes; Cloud agents continue until sandbox sleep. The page still documents shared Cloud workspaces and setup.

Open-source license: No. Proprietary

Open-source license

Typed value: No. Proprietary.

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Client license: Proprietary vendor software/service terms; source-visible integrations do not license the main application.
  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

Any desktop app: Yes

Any desktop app

Typed value: Yes.

Conductor documents a macOS app and an iPhone app requiring iOS 26.0+. Release 0.92.0 adds experimental remote control of local Mac workspaces from iPhone. Current installation instructions provide Mac download steps; the announcement no longer states a signup closure.

  • Windows/Linux are explicitly unavailable; Android remains undocumented.
  • Remote Control requires the Mac and experimental setting. No app or authenticated Cloud test was run.
  1. Install Conductor

    Finding: The install guide provides Mac download/setup steps, explicitly excludes Windows/Linux, and requires GitHub and an agent provider. It does not state an existing-account-only restriction.

All public findings

Findings may summarize a source. They are not presented as verbatim statements.

conductor:ACC-01: Conductor documents a macOS app and an iPhone app requiring iOS 26.0+. Release 0.92.0 adds experimental remote control of local Mac workspaces from iPhone. Current installation instructions provide Mac download steps; the announcement no longer states a signup closure.

conductor:ACC-01

Conductor documents a macOS app and an iPhone app requiring iOS 26.0+. Release 0.92.0 adds experimental remote control of local Mac workspaces from iPhone. Current installation instructions provide Mac download steps; the announcement no longer states a signup closure.

  • Windows/Linux are explicitly unavailable; Android remains undocumented.
  • Remote Control requires the Mac and experimental setting. No app or authenticated Cloud test was run.
  1. Install

    Finding: macOS only (not available for Windows or Linux yet); setup checks GitHub auth and at least one of Claude, Codex or Cursor credentials.

  2. Conductor

    Finding: Homepage describes Cloud workspaces, multiplayer, four harnesses and mobile/API access. Latest changelog card is 0.93.0 dated November 4, 2026, joining Meta; acquisition banner links the announcement.

  3. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  4. Conductor API

    Finding: Beta HTTP API under https://api.conductor.build/v0 with an OpenAPI document; bearer API key; endpoints for projects, workspaces (create, get, rename, archive, unarchive, sleep), sessions, prompts, transcripts, cancel ("drop queued messages"), SQL and routines. Routines created through the API run a saved prompt in a fresh workspace on each webhook JSON POST. Cookbook: "messages sent while it works are steered into the running turn"; Tips: a queued prompt reports idle until its turn starts. Some 403s come from a proxy that rejects default client signatures (send a User-Agent). Shows a routine example with the `conductor` CLI. No SDK is documented on this page. Quick start (for AI) requires a Conductor Pro account to generate an API key. Page undated. Endpoint table: GET /v0/sessions/{id}/status "Whether the agent is idle, working, or errored"; POST /v0/sessions/{id}/cancel "Stop the agent's current turn and drop queued messages". Tips: "Always pass an explicit model" (workspace creation, POST /v0/workspaces) and "An optional effort field sets the thinking level for claude and codex sessions". No sidebar-section endpoint listed.

  5. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  6. Conductor changelog

    Finding: Latest entry 0.93.0 dated November 4, 2026 announces Meta and wind-down. 0.92.0 dated October 23 describes iPhone remote control of local Mac workspaces and a generally enabled Artifacts tab. Historical entries remain available.

  7. Conductor 0.90.0: Conductor for iOS

    Finding: iOS app (App Store link id6791228564): chat with agents and merge PRs from a phone, works with cloud workspaces. In cloud organizations pinned workspaces sync across devices and Automerge, optional-check and PR-order settings carry over to iOS. Notifications show the chat title instead of the branch name (no client named). Misc: "Pausing, resuming, or clearing a Codex goal shows up in the chat." Undated "New in 0.90.1" notes: background-command chats show as waiting and queued messages still send. No plan or iOS version requirement stated.

  8. conductor-releases issues

    Finding: All 29 open issues read on 2026-10-09 (newest #35, 2026-10-09). Feature requests include secure mobile/remote control of local sessions (#22, 2026-07-12), Claude Code /remote-control (#19), SSH remote development (#14), multiple GitHub accounts (#33), Go to Definition (#30) and per-workspace notes (#35). #32 (opened 2026-09-19, open, no comments) reports an API task message returning HTTP 403 "Blocked". No maintainer comments, labels or announced plans seen.

  9. Conductor 0.77.0

    Finding: Alpha multiplayer and API for Pro users; "Workspace notifications now show the workspace name instead of the branch name" (no client named); experimental cloud cost estimates include Codex usage; background tasks live for everyone. API alpha: "Available to all Pro users." No price stated.

  10. Conductor 0.78.0: Introducing Conductor Cloud

    Finding: Cloud workspaces on isolated microVMs that keep running when the laptop closes; start by signing up for Pro and creating an organization; shareable links, presence and teammates prompting agents together in real time; API to manage cloud workspaces. Said the mobile app was "coming soon" on 2026-07-30. "To get started, sign up for Pro and create an organization." Billing "opens the Stripe billing portal". No price stated.

  11. Conductor 0.80.0: Stacks

    Finding: GitHub stacks (cloud; local needs gh stack extension); cloud workspaces include the Conductor CLI and Chrome; cloud environment variables moved to Organization → Environment and "editing them no longer requires rebuilding your Cloud Computer"; GH_HOST respected for GitHub Enterprise.

  12. Conductor 0.83.0: Auto port forwarding

    Finding: Cloud auto port forwarding; one-click PR approval from the git panel; messages to sleeping cloud workspaces wake them; View deployment(s) button; added completion sounds; "Fixed notifications for workspaces in another organization opening the dashboard instead of the chat" (no client named).

  13. Conductor 0.87.0: Search across files, OpenCode 2.0

    Finding: Routine GitHub trigger filters (label, author, base branch) and weekday schedules; OpenCode 2.0; "Create and delete sidebar sections from the CLI, API, and MCP in cloud organizations"; with the experimental status matrix enabled, indicators "show distinct states for routine runs, attention, errors, running pull request checks, and merged pull requests"; video thumbnails in chat; security fixes; setup scripts from shared cloud workspaces no longer auto-run on collaborators' Macs.

  14. Conductor Remote Control

    Finding: Experimental feature: the Conductor iOS app pairs with a Mac by QR code and can browse and create local workspaces, read chats, send messages, pick models, stop agents and archive workspaces. Works with local workspaces only; the Mac app must be open, awake and online, both devices on the same account; off while enterprise data privacy is on; traffic is end-to-end encrypted through Conductor's relay.

  15. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  16. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

  17. Install Conductor

    Finding: The install guide provides Mac download/setup steps, explicitly excludes Windows/Linux, and requires GitHub and an agent provider. It does not state an existing-account-only restriction.

  18. Conductor.build App Store listing

    Finding: The US App Store lists the Melty iPhone app as free to download, requiring iOS 26.0 or later. The displayed version is 1.1.3. Store download price does not establish Conductor plan entitlements or successful account creation.

conductor:ACC-02: Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled. cross-checked

conductor:ACC-02

Four agent harnesses (Claude Code, Codex, Cursor, OpenCode) with several model-provider routes; Claude Code, Codex and OpenCode are bundled.

  • Cursor runs through the Cursor API rather than a local executable.
  • Provider coverage per harness and in Cloud was not tested; Vertex AI is named only as a Claude provider setting (settings reference) with no setup guide located, and custom harness adapters are not documented.
  • Mixed-harness operation is documented but was not observed.
  1. Conductor

    Finding: Vendor homepage presents a team of coding agents running in the cloud, isolated cloud workspaces, and shared workspaces where teammates can follow work and prompt together in real time. It names Claude Code, Codex, Cursor and OpenCode. The Conduct from anywhere section links macOS, the iOS App Store and the API; the latest release link is 0.90.0 (Conductor for iOS, 2026-10-02).

  2. Parallel agents

    Finding: Independent branches or multiple agents sharing one checkout; shared-file collision risk is explicit.

  3. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

  4. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

  5. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  6. Settings reference

    Finding: Settings include tool_approvals_enabled, Codex tool_approval_policy (default or always_approve), default plan mode, review/PR/conflict prompts, git.archive_on_merge, enterprise_data_privacy.

conductor:ACC-03: Subscriptions and API keys are accepted locally and in Cloud; Sign in with ChatGPT was added for Codex; Conductor does not resell model usage. cross-checked

conductor:ACC-03

Subscriptions and API keys are accepted locally and in Cloud; Sign in with ChatGPT was added for Codex; Conductor does not resell model usage.

  • Vendor credits and provider restrictions remain unestablished; the OpenCode local-model route is documented.
  • Credential flows were not exercised.
  1. Install

    Finding: macOS only (not available for Windows or Linux yet); setup checks GitHub auth and at least one of Claude, Codex or Cursor credentials.

  2. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  3. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  4. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

  5. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  6. Set up cloud

    Finding: Cloud onboarding: create organization, connect the Conductor GitHub App, add an agent credential (subscription token or API key), build the Cloud Computer, then work in an Admin workspace. Admins can add shared API keys that take precedence over personal keys.

  7. FAQ

    Finding: Agents run unsandboxed with the user account access; auth reuses Claude Code/Codex logins with an API-key mode. "How does Conductor make money? Right now we don't." and "At some point we plan to charge for collaboration features". Moving computers: "Conductor stores files at ~/Library/Application Support/com.conductor.app and ~/conductor." Page undated.

  8. Conductor 0.84.0: Edit files directly in diffs

    Finding: Edit files in diffs; effort change mid-chat for some Claude models; "Codex chats now retry automatically when OpenAI rate-limits a request" (no surface named); "Port forwarding is now off by default for each workspace"; PR links can open in GitHub or Linear (Settings → Git); 0.84.1 swaps cloud agent auth without restart; remote desktop control in cloud (experimental).

  9. Conductor 0.89.0: Sign in with ChatGPT

    Finding: Sign in with ChatGPT for Codex; "Steer Cursor chats while they're running, in local and cloud workspaces."; PR view top card; merge blocked if the PR changed on GitHub; Codex tasks no longer stop on temporary rate-limit errors; custom OAuth MCP servers can use the Conductor callback URL. PR view: "open the Review row to add unresolved comments to chat, or the Checks row to re-run or fix failed checks."

conductor:ACC-04: Git-backed setup with credential checks; GitHub required; Cloud onboarding is guided and BYO cloud machines are unavailable. cross-checked

conductor:ACC-04

Git-backed setup with credential checks; GitHub required; Cloud onboarding is guided and BYO cloud machines are unavailable.

  • Whether a local repository without GitHub works fully is not documented; install docs require GitHub authentication.
  • Cloud supports GitHub repositories only.
  1. Install

    Finding: macOS only (not available for Windows or Linux yet); setup checks GitHub auth and at least one of Claude, Codex or Cursor credentials.

  2. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  3. First workspace

    Finding: Repository options: "Open project" (existing local repository), "Open GitHub project" and "Quick start" (new repository). Setup and run scripts; chats with Claude Code, Codex, Cursor or OpenCode; gh auth status check. No text on repositories without a remote or non-git folders. Page undated.

  4. Cloud Computer

    Finding: Each cloud workspace runs in an isolated microVM with 8 vCPU, 16 GB RAM and 32 GB ephemeral NVMe. The Cloud Computer is the shared development environment for a Conductor organization; the page gives no Cloud Computer disk size; up to 50 GitHub repositories; successful builds seed new workspaces; saved secrets are hidden but exposed as env vars inside workspaces; an Admin workspace agent can edit install/setup scripts and run builds.

  5. Set up cloud

    Finding: Cloud onboarding: create organization, connect the Conductor GitHub App, add an agent credential (subscription token or API key), build the Cloud Computer, then work in an Admin workspace. Admins can add shared API keys that take precedence over personal keys.

  6. Conductor 0.80.0: Stacks

    Finding: GitHub stacks (cloud; local needs gh stack extension); cloud workspaces include the Conductor CLI and Chrome; cloud environment variables moved to Organization → Environment and "editing them no longer requires rebuilding your Cloud Computer"; GH_HOST respected for GitHub Enterprise.

conductor:ACC-05: Beta HTTP API (/v0, OpenAPI), hosted Streamable HTTP MCP server and routine webhooks; Conductor CLI exists in cloud workspaces. cross-checked

conductor:ACC-05

Beta HTTP API (/v0, OpenAPI), hosted Streamable HTTP MCP server and routine webhooks; Conductor CLI exists in cloud workspaces.

  • API and MCP are labelled beta and may change.
  • No SDK located. Public issue #32 (opened 2026-09-19, open, no comments) reports an API task message returning HTTP 403 "Blocked". The API doc attributes some 403s to a proxy that rejects default client signatures and advises sending a User-Agent. Not reproduced here.
  • The API Quick start requires a Conductor Pro account.
  1. Conductor API

    Finding: Beta HTTP API under https://api.conductor.build/v0 with an OpenAPI document; bearer API key; endpoints for projects, workspaces (create, get, rename, archive, unarchive, sleep), sessions, prompts, transcripts, cancel ("drop queued messages"), SQL and routines. Routines created through the API run a saved prompt in a fresh workspace on each webhook JSON POST. Cookbook: "messages sent while it works are steered into the running turn"; Tips: a queued prompt reports idle until its turn starts. Some 403s come from a proxy that rejects default client signatures (send a User-Agent). Shows a routine example with the `conductor` CLI. No SDK is documented on this page. Quick start (for AI) requires a Conductor Pro account to generate an API key. Page undated. Endpoint table: GET /v0/sessions/{id}/status "Whether the agent is idle, working, or errored"; POST /v0/sessions/{id}/cancel "Stop the agent's current turn and drop queued messages". Tips: "Always pass an explicit model" (workspace creation, POST /v0/workspaces) and "An optional effort field sets the thinking level for claude and codex sessions". No sidebar-section endpoint listed.

  2. Conductor MCP server

    Finding: Beta hosted MCP ("Tools and input shapes may change"): OAuth limited to mcp:tools and one organization, or bearer API key. Tools include create_session, send_message ("Send or queue a prompt for a session"), cancel_session ("Cancel the current agent turn and discard queued messages"), get_session_status ("Get a session's current agent status", no values listed), list_models, list_messages (after cursor), get_message and run_sql. No routine or section tools listed. Page undated.

  3. conductor-releases issues

    Finding: All 29 open issues read on 2026-10-09 (newest #35, 2026-10-09). Feature requests include secure mobile/remote control of local sessions (#22, 2026-07-12), Claude Code /remote-control (#19), SSH remote development (#14), multiple GitHub accounts (#33), Go to Definition (#30) and per-workspace notes (#35). #32 (opened 2026-09-19, open, no comments) reports an API task message returning HTTP 403 "Blocked". No maintainer comments, labels or announced plans seen.

  4. Conductor 0.80.0: Stacks

    Finding: GitHub stacks (cloud; local needs gh stack extension); cloud workspaces include the Conductor CLI and Chrome; cloud environment variables moved to Organization → Environment and "editing them no longer requires rebuilding your Cloud Computer"; GH_HOST respected for GitHub Enterprise.

  5. Conductor 0.87.0: Search across files, OpenCode 2.0

    Finding: Routine GitHub trigger filters (label, author, base branch) and weekday schedules; OpenCode 2.0; "Create and delete sidebar sections from the CLI, API, and MCP in cloud organizations"; with the experimental status matrix enabled, indicators "show distinct states for routine runs, attention, errors, running pull request checks, and merged pull requests"; video thumbnails in chat; security fixes; setup scripts from shared cloud workspaces no longer auto-run on collaborators' Macs.

conductor:ACC-06: MCP client use through each harness with user/project scopes, OAuth and approval prompts; instruction files and skills are read by agents. cross-checked

conductor:ACC-06

MCP client use through each harness with user/project scopes, OAuth and approval prompts; instruction files and skills are read by agents.

  • Tool approvals vary by agent and setting; Cursor MCP status is not shown in Conductor.
  • No plugin system located.
  1. Security and permissions

    Finding: Cloud agents run in isolated Linux sandboxes hosted by Vercel and can use the workspace environment variables and secrets; repository files and session data are kept in Conductor-managed infrastructure. Local agents run with the user account permissions; approvals can be required for shell commands, file changes, MCP tool calls or web fetches. Page undated.

  2. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  3. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

  4. Privacy

    Finding: Cloud session inputs and outputs are stored on Conductor servers; local session data stays on the Mac (~/Library/Application Support/com.conductor.app); analytics and crash logs via PostHog, including OS and IP address. No export, deletion or retention text. Enterprise data privacy is local-only and "cannot be used in cloud". Page undated.

  5. Settings reference

    Finding: Settings include tool_approvals_enabled, Codex tool_approval_policy (default or always_approve), default plan mode, review/PR/conflict prompts, git.archive_on_merge, enterprise_data_privacy.

  6. Configure MCP servers

    Finding: MCP servers are configured in each harness scope (user/project) and loaded by the agent host; status UI covers Claude Code and Codex, not Cursor; OAuth via status row; approval prompts depend on agent settings.

  7. MCP

    Finding: MCP works where the agent host supports it; transports stdio, Streamable HTTP, SSE; user/project/local scopes; enterprise data privacy disables custom MCP servers.

  8. Agent behavior

    Finding: Session controls table: Plan Mode, Fast Mode ("where supported"), thinking or reasoning level, Codex personality, checkpoints; no per-agent support listed. Repository preferences for review, PR, error-fix, conflict resolution and branch rename; agents read AGENTS.md, CLAUDE.md, .claude/commands and skills.

  9. Conductor 0.82.0: Conductor MCP

    Finding: Conductor MCP launched; cloud MCP servers; "Your Cloud Computer got more oomph - they now have double the disk space (64GB)"; low-disk warnings.

  10. Conductor 0.89.0: Sign in with ChatGPT

    Finding: Sign in with ChatGPT for Codex; "Steer Cursor chats while they're running, in local and cloud workspaces."; PR view top card; merge blocked if the PR changed on GitHub; Codex tasks no longer stop on temporary rate-limit errors; custom OAuth MCP servers can use the Conductor callback URL. PR view: "open the Review row to add unresolved comments to chat, or the Checks row to re-run or fix failed checks."

conductor:PLAN-01: GitHub and Linear issues, branches and PRs can start workspaces; prompt links prefill the composer. cross-checked

conductor:PLAN-01

GitHub and Linear issues, branches and PRs can start workspaces; prompt links prefill the composer.

  • The issue-to-PR guide does not describe linking the issue to the resulting PR; sync and round-trips are not documented.
  1. Workflow

    Finding: Workspace is the delegation unit; branch/PR is integration unit. GitHub/Linear intake and repository instructions are documented.

  2. First workspace

    Finding: Repository options: "Open project" (existing local repository), "Open GitHub project" and "Quick start" (new repository). Setup and run scripts; chats with Claude Code, Codex, Cursor or OpenCode; gh auth status check. No text on repositories without a remote or non-git folders. Page undated.

  3. From issue to pull request

    Finding: A workspace can be started from a GitHub or Linear issue and inherits its title, description and context; Create PR sends the diff and repository context to the agent to draft the PR. Issue-PR linking is not described. After the PR exists, use Checks and Diff Viewer "to respond to CI, deployments, review comments, and todos".

  4. Conductor 0.86.0: Share your loadouts

    Finding: "Removed Codex goals and message queuing. Messages sent while an agent is working now steer it instead of waiting in a queue" (no surface named); answer Codex questions while it works (no surface named); context usage popover shows Claude account limits and reset times, matching Codex; cloud run scripts; private workspaces marked with a lock; fixes for OpenCode rejecting Always allow approvals and for "cloud plan approvals appearing successful when the workspace rejected the response". Improvements also: "Open pull requests in Graphite"; "Prompt links fill the new-workspace composer in your active organization." The Always allow fix is in the Fixes list.

conductor:PLAN-02: Plan Mode for Claude and Codex and workspace todos organize longer work. Plans saved in .context sync from Cloud to the Mac; release 0.90.0 restores the Approve control for a pending plan after restart. cross-checked

conductor:PLAN-02

Plan Mode for Claude and Codex and workspace todos organize longer work. Plans saved in .context sync from Cloud to the Mac; release 0.90.0 restores the Approve control for a pending plan after restart.

  • Plan revision history and complete plan/todo durability guarantees remain unestablished.
  1. Todos

    Finding: Humans and agents use workspace todos; unchecked todos block merge; @todos exposes list to agent.

  2. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  3. Conductor 0.90.0: Conductor for iOS

    Finding: iOS app (App Store link id6791228564): chat with agents and merge PRs from a phone, works with cloud workspaces. In cloud organizations pinned workspaces sync across devices and Automerge, optional-check and PR-order settings carry over to iOS. Notifications show the chat title instead of the branch name (no client named). Misc: "Pausing, resuming, or clearing a Codex goal shows up in the chat." Undated "New in 0.90.1" notes: background-command chats show as waiting and queued messages still send. No plan or iOS version requirement stated.

  4. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

conductor:PLAN-03: Local workspaces have todos that block merging until checked off. Epics, hierarchy, dependency graphs and state propagation were not established; no installed-app test was run.

conductor:PLAN-03

Local workspaces have todos that block merging until checked off. Epics, hierarchy, dependency graphs and state propagation were not established; no installed-app test was run.

  • Todos are a per-workspace merge checklist in the local app, not an issue dependency graph; epics, hierarchy and state propagation were not established.
  • Documentation only; no installed-app test was run.
  1. Conductor todos

    Finding: People and agents add workspace todos. Open todos block merging; @todos sends the list to the agent.

conductor:PLAN-04: Repository settings, instruction files and skills carry shared guidance; persistent memory is not documented. cross-checked

conductor:PLAN-04

Repository settings, instruction files and skills carry shared guidance; persistent memory is not documented.

  • Memory, provenance and correction are not documented.
  1. Workflow

    Finding: Workspace is the delegation unit; branch/PR is integration unit. GitHub/Linear intake and repository instructions are documented.

  2. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

  3. Isolated workspaces

    Finding: Each workspace is a git worktree on its own branch; a branch can be checked out in one workspace at a time; states workspace isolation is development isolation, not a security boundary, and agents run on the Mac with user permissions.

  4. Agent behavior

    Finding: Session controls table: Plan Mode, Fast Mode ("where supported"), thinking or reasoning level, Codex personality, checkpoints; no per-agent support listed. Repository preferences for review, PR, error-fix, conflict resolution and branch rename; agents read AGENTS.md, CLAUDE.md, .claude/commands and skills.

conductor:PLAN-05: Attachments of notes, specs, screenshots and logs are supported, and file-content search exists (0.87.0); indexing, retrieval and citations are not documented. cross-checked

conductor:PLAN-05

Attachments of notes, specs, screenshots and logs are supported, and file-content search exists (0.87.0); indexing, retrieval and citations are not documented.

  • Only attachments and file-content search are documented; no indexing, retrieval or citation features located.
  1. First workspace

    Finding: Repository options: "Open project" (existing local repository), "Open GitHub project" and "Quick start" (new repository). Setup and run scripts; chats with Claude Code, Codex, Cursor or OpenCode; gh auth status check. No text on repositories without a remote or non-git folders. Page undated.

  2. Conductor 0.82.0: Conductor MCP

    Finding: Conductor MCP launched; cloud MCP servers; "Your Cloud Computer got more oomph - they now have double the disk space (64GB)"; low-disk warnings.

  3. Conductor 0.87.0: Search across files, OpenCode 2.0

    Finding: Routine GitHub trigger filters (label, author, base branch) and weekday schedules; OpenCode 2.0; "Create and delete sidebar sections from the CLI, API, and MCP in cloud organizations"; with the experimental status matrix enabled, indicators "show distinct states for routine runs, attention, errors, running pull request checks, and merged pull requests"; video thumbnails in chat; security fixes; setup scripts from shared cloud workspaces no longer auto-run on collaborators' Macs.

conductor:PLAN-06: Workspace todos exist; an independent issue backlog is unverified.

conductor:PLAN-06

Workspace todos exist; an independent issue backlog is unverified.

  • Cannot conclude native issues are absent. Parking before a workspace, hierarchy and rich metadata need direct research.
  1. Workflow

    Finding: Workspace is the delegation unit; branch/PR is integration unit. GitHub/Linear intake and repository instructions are documented.

  2. Todos

    Finding: Humans and agents use workspace todos; unchecked todos block merge; @todos exposes list to agent.

  3. Conductor todos

    Finding: People and agents add workspace todos. Open todos block merging; @todos sends the list to the agent.

conductor:PLAN-07: Agents can receive workspace todos; shared native issue operations remain unverified. cross-checked

conductor:PLAN-07

Agents can receive workspace todos; shared native issue operations remain unverified.

  • The MCP creates and manages workspaces/sessions; those operations do not establish native issue CRUD, shared issue IDs, dependencies or issue claims. No shared human/agent issue scenario was run.
  • Release 0.87.0 documents sidebar-section operations through CLI/API/MCP. Omission from the undated endpoint/tool tables is a documentation gap, not contrary evidence; sidebar sections do not establish native issue operations.
  1. Todos

    Finding: Humans and agents use workspace todos; unchecked todos block merge; @todos exposes list to agent.

  2. Conductor API

    Finding: Beta HTTP API under https://api.conductor.build/v0 with an OpenAPI document; bearer API key; endpoints for projects, workspaces (create, get, rename, archive, unarchive, sleep), sessions, prompts, transcripts, cancel ("drop queued messages"), SQL and routines. Routines created through the API run a saved prompt in a fresh workspace on each webhook JSON POST. Cookbook: "messages sent while it works are steered into the running turn"; Tips: a queued prompt reports idle until its turn starts. Some 403s come from a proxy that rejects default client signatures (send a User-Agent). Shows a routine example with the `conductor` CLI. No SDK is documented on this page. Quick start (for AI) requires a Conductor Pro account to generate an API key. Page undated. Endpoint table: GET /v0/sessions/{id}/status "Whether the agent is idle, working, or errored"; POST /v0/sessions/{id}/cancel "Stop the agent's current turn and drop queued messages". Tips: "Always pass an explicit model" (workspace creation, POST /v0/workspaces) and "An optional effort field sets the thinking level for claude and codex sessions". No sidebar-section endpoint listed.

  3. Conductor MCP server

    Finding: Beta hosted MCP ("Tools and input shapes may change"): OAuth limited to mcp:tools and one organization, or bearer API key. Tools include create_session, send_message ("Send or queue a prompt for a session"), cancel_session ("Cancel the current agent turn and discard queued messages"), get_session_status ("Get a session's current agent status", no values listed), list_models, list_messages (after cursor), get_message and run_sql. No routine or section tools listed. Page undated.

  4. Conductor 0.87.0: Search across files, OpenCode 2.0

    Finding: Routine GitHub trigger filters (label, author, base branch) and weekday schedules; OpenCode 2.0; "Create and delete sidebar sections from the CLI, API, and MCP in cloud organizations"; with the experimental status matrix enabled, indicators "show distinct states for routine runs, attention, errors, running pull request checks, and merged pull requests"; video thumbnails in chat; security fixes; setup scripts from shared cloud workspaces no longer auto-run on collaborators' Macs.

conductor:COORD-01: Multiple workspaces, or several chats in one workspace, run in parallel; Home and the sidebar show workspace status, and a matrix-style status indicator was introduced as experimental opt-in (0.85.0) and is shown in cloud organizations per 0.90.0. cross-checked

conductor:COORD-01

Multiple workspaces, or several chats in one workspace, run in parallel; Home and the sidebar show workspace status, and a matrix-style status indicator was introduced as experimental opt-in (0.85.0) and is shown in cloud organizations per 0.90.0.

  • Concurrency quotas and large-fleet behavior were not tested; same-workspace agents can edit the same files.
  1. Parallel agents

    Finding: Independent branches or multiple agents sharing one checkout; shared-file collision risk is explicit.

  2. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  3. Conductor 0.90.0: Conductor for iOS

    Finding: iOS app (App Store link id6791228564): chat with agents and merge PRs from a phone, works with cloud workspaces. In cloud organizations pinned workspaces sync across devices and Automerge, optional-check and PR-order settings carry over to iOS. Notifications show the chat title instead of the branch name (no client named). Misc: "Pausing, resuming, or clearing a Codex goal shows up in the chat." Undated "New in 0.90.1" notes: background-command chats show as waiting and queued messages still send. No plan or iOS version requirement stated.

  4. Conductor 0.87.0: Search across files, OpenCode 2.0

    Finding: Routine GitHub trigger filters (label, author, base branch) and weekday schedules; OpenCode 2.0; "Create and delete sidebar sections from the CLI, API, and MCP in cloud organizations"; with the experimental status matrix enabled, indicators "show distinct states for routine runs, attention, errors, running pull request checks, and merged pull requests"; video thumbnails in chat; security fixes; setup scripts from shared cloud workspaces no longer auto-run on collaborators' Macs.

conductor:COORD-02: Cloud workspaces can be reassigned to teammates; agent assignment is not documented. cross-checked

conductor:COORD-02

Cloud workspaces can be reassigned to teammates; agent assignment is not documented.

  • Assignment is manual; routing and claims are not documented.
  1. Multiplayer

    Finding: Cloud organization workspaces are shared with the team; transcripts update live; presence dots for active teammates and typing indicators in a shared chat; Home filters by teammate and project; follow and reassign workspaces; direct links. The page does not describe co-prompting, private workspaces, roles or plan limits. Page undated.

conductor:COORD-03: API/MCP (beta, cloud workspaces) can create sessions, send prompts, cancel the current turn and read transcripts; an external MCP client can orchestrate Conductor sessions. The API doc says messages sent while an agent works are steered into the running turn; the MCP reference still describes send_message as "send or queue". cross-checked

conductor:COORD-03

API/MCP (beta, cloud workspaces) can create sessions, send prompts, cancel the current turn and read transcripts; an external MCP client can orchestrate Conductor sessions. The API doc says messages sent while an agent works are steered into the running turn; the MCP reference still describes send_message as "send or queue".

  • Cloud workspaces only; beta interfaces.
  • HTTP mid-turn steering is documented; MCP mid-turn handling is unspecified by its send-or-queue description. No runtime test was performed.
  1. Conductor API

    Finding: Beta HTTP API under https://api.conductor.build/v0 with an OpenAPI document; bearer API key; endpoints for projects, workspaces (create, get, rename, archive, unarchive, sleep), sessions, prompts, transcripts, cancel ("drop queued messages"), SQL and routines. Routines created through the API run a saved prompt in a fresh workspace on each webhook JSON POST. Cookbook: "messages sent while it works are steered into the running turn"; Tips: a queued prompt reports idle until its turn starts. Some 403s come from a proxy that rejects default client signatures (send a User-Agent). Shows a routine example with the `conductor` CLI. No SDK is documented on this page. Quick start (for AI) requires a Conductor Pro account to generate an API key. Page undated. Endpoint table: GET /v0/sessions/{id}/status "Whether the agent is idle, working, or errored"; POST /v0/sessions/{id}/cancel "Stop the agent's current turn and drop queued messages". Tips: "Always pass an explicit model" (workspace creation, POST /v0/workspaces) and "An optional effort field sets the thinking level for claude and codex sessions". No sidebar-section endpoint listed.

  2. Conductor MCP server

    Finding: Beta hosted MCP ("Tools and input shapes may change"): OAuth limited to mcp:tools and one organization, or bearer API key. Tools include create_session, send_message ("Send or queue a prompt for a session"), cancel_session ("Cancel the current agent turn and discard queued messages"), get_session_status ("Get a session's current agent status", no values listed), list_models, list_messages (after cursor), get_message and run_sql. No routine or section tools listed. Page undated.

  3. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  4. Conductor 0.86.0: Share your loadouts

    Finding: "Removed Codex goals and message queuing. Messages sent while an agent is working now steer it instead of waiting in a queue" (no surface named); answer Codex questions while it works (no surface named); context usage popover shows Claude account limits and reset times, matching Codex; cloud run scripts; private workspaces marked with a lock; fixes for OpenCode rejecting Always allow approvals and for "cloud plan approvals appearing successful when the workspace rejected the response". Improvements also: "Open pull requests in Graphite"; "Prompt links fill the new-workspace composer in your active organization." The Always allow fix is in the Fixes list.

conductor:COORD-04: Cloud routines run saved prompts on schedules (including weekdays), GitHub actions with filters, or webhooks; routines are Pro/Team only. cross-checked

conductor:COORD-04

Cloud routines run saved prompts on schedules (including weekdays), GitHub actions with filters, or webhooks; routines are Pro/Team only.

  • Overlap, retry and completion behavior are not documented.
  • Routines require Pro/Team and run only in Cloud.
  1. Conductor Cloud

    Finding: Each agent runs in its own isolated cloud sandbox and keeps working after the laptop closes until the sandbox sleeps; "The chat, diff, git, terminal, and pull request workflow is the same in both" local and cloud. Support via Help > Send Feedback or humans@conductor.build. Page undated. Local vs Cloud table: "After your Mac closes | Local processes stop | Agents keep running until the sandbox sleeps".

  2. Conductor API

    Finding: Beta HTTP API under https://api.conductor.build/v0 with an OpenAPI document; bearer API key; endpoints for projects, workspaces (create, get, rename, archive, unarchive, sleep), sessions, prompts, transcripts, cancel ("drop queued messages"), SQL and routines. Routines created through the API run a saved prompt in a fresh workspace on each webhook JSON POST. Cookbook: "messages sent while it works are steered into the running turn"; Tips: a queued prompt reports idle until its turn starts. Some 403s come from a proxy that rejects default client signatures (send a User-Agent). Shows a routine example with the `conductor` CLI. No SDK is documented on this page. Quick start (for AI) requires a Conductor Pro account to generate an API key. Page undated. Endpoint table: GET /v0/sessions/{id}/status "Whether the agent is idle, working, or errored"; POST /v0/sessions/{id}/cancel "Stop the agent's current turn and drop queued messages". Tips: "Always pass an explicit model" (workspace creation, POST /v0/workspaces) and "An optional effort field sets the thinking level for claude and codex sessions". No sidebar-section endpoint listed.

  3. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  4. Conductor 0.87.0: Search across files, OpenCode 2.0

    Finding: Routine GitHub trigger filters (label, author, base branch) and weekday schedules; OpenCode 2.0; "Create and delete sidebar sections from the CLI, API, and MCP in cloud organizations"; with the experimental status matrix enabled, indicators "show distinct states for routine runs, attention, errors, running pull request checks, and merged pull requests"; video thumbnails in chat; security fixes; setup scripts from shared cloud workspaces no longer auto-run on collaborators' Macs.

conductor:COORD-05: Cloud scope. Stop: cancel_session (MCP) and POST /v0/sessions/{id}/cancel end the current turn and drop queued messages. Steer: changelog 0.89.0 (2026-09-29) says Cursor chats can be steered while running in local and cloud workspaces, and the API doc says messages sent while an agent works are steered into the running turn; the MCP reference still says send_message "Send or queue a prompt for a session". Pending decisions: 0.86.0 fixed cloud plan approvals, so plan approvals happen in cloud workspaces. Pause and resume are not established for cloud, and notifications for cloud runs are not documented. Not tested hands-on. cross-checked

conductor:COORD-05

Cloud scope. Stop: cancel_session (MCP) and POST /v0/sessions/{id}/cancel end the current turn and drop queued messages. Steer: changelog 0.89.0 (2026-09-29) says Cursor chats can be steered while running in local and cloud workspaces, and the API doc says messages sent while an agent works are steered into the running turn; the MCP reference still says send_message "Send or queue a prompt for a session". Pending decisions: 0.86.0 fixed cloud plan approvals, so plan approvals happen in cloud workspaces. Pause and resume are not established for cloud, and notifications for cloud runs are not documented. Not tested hands-on.

  • Intervention behavior was not tested.
  • Cursor steering in local and cloud workspaces is documented in 0.89.0, and HTTP steering in the API cookbook. MCP send-or-queue wording does not establish its mid-turn behavior or contradict the HTTP description. Other Mac-app harness steering remains unverified.
  • Notifications and Codex question handling are documented without a surface and are not claimed for cloud.
  • Codex goals (pause/resume) are documented for local workspaces only and are scoped out of this cloud claim. Changelog 0.86.0 said "Removed Codex goals and message queuing"; 0.90.0 (2026-10-02) says "Pausing, resuming, or clearing a Codex goal shows up in the chat" without naming a surface.
  1. Conductor API

    Finding: Beta HTTP API under https://api.conductor.build/v0 with an OpenAPI document; bearer API key; endpoints for projects, workspaces (create, get, rename, archive, unarchive, sleep), sessions, prompts, transcripts, cancel ("drop queued messages"), SQL and routines. Routines created through the API run a saved prompt in a fresh workspace on each webhook JSON POST. Cookbook: "messages sent while it works are steered into the running turn"; Tips: a queued prompt reports idle until its turn starts. Some 403s come from a proxy that rejects default client signatures (send a User-Agent). Shows a routine example with the `conductor` CLI. No SDK is documented on this page. Quick start (for AI) requires a Conductor Pro account to generate an API key. Page undated. Endpoint table: GET /v0/sessions/{id}/status "Whether the agent is idle, working, or errored"; POST /v0/sessions/{id}/cancel "Stop the agent's current turn and drop queued messages". Tips: "Always pass an explicit model" (workspace creation, POST /v0/workspaces) and "An optional effort field sets the thinking level for claude and codex sessions". No sidebar-section endpoint listed.

  2. Conductor MCP server

    Finding: Beta hosted MCP ("Tools and input shapes may change"): OAuth limited to mcp:tools and one organization, or bearer API key. Tools include create_session, send_message ("Send or queue a prompt for a session"), cancel_session ("Cancel the current agent turn and discard queued messages"), get_session_status ("Get a session's current agent status", no values listed), list_models, list_messages (after cursor), get_message and run_sql. No routine or section tools listed. Page undated.

  3. Conductor changelog

    Finding: Latest entry 0.93.0 dated November 4, 2026 announces Meta and wind-down. 0.92.0 dated October 23 describes iPhone remote control of local Mac workspaces and a generally enabled Artifacts tab. Historical entries remain available.

  4. Conductor 0.90.0: Conductor for iOS

    Finding: iOS app (App Store link id6791228564): chat with agents and merge PRs from a phone, works with cloud workspaces. In cloud organizations pinned workspaces sync across devices and Automerge, optional-check and PR-order settings carry over to iOS. Notifications show the chat title instead of the branch name (no client named). Misc: "Pausing, resuming, or clearing a Codex goal shows up in the chat." Undated "New in 0.90.1" notes: background-command chats show as waiting and queued messages still send. No plan or iOS version requirement stated.

  5. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

  6. Conductor 0.77.0

    Finding: Alpha multiplayer and API for Pro users; "Workspace notifications now show the workspace name instead of the branch name" (no client named); experimental cloud cost estimates include Codex usage; background tasks live for everyone. API alpha: "Available to all Pro users." No price stated.

  7. Conductor 0.83.0: Auto port forwarding

    Finding: Cloud auto port forwarding; one-click PR approval from the git panel; messages to sleeping cloud workspaces wake them; View deployment(s) button; added completion sounds; "Fixed notifications for workspaces in another organization opening the dashboard instead of the chat" (no client named).

  8. Conductor 0.86.0: Share your loadouts

    Finding: "Removed Codex goals and message queuing. Messages sent while an agent is working now steer it instead of waiting in a queue" (no surface named); answer Codex questions while it works (no surface named); context usage popover shows Claude account limits and reset times, matching Codex; cloud run scripts; private workspaces marked with a lock; fixes for OpenCode rejecting Always allow approvals and for "cloud plan approvals appearing successful when the workspace rejected the response". Improvements also: "Open pull requests in Graphite"; "Prompt links fill the new-workspace composer in your active organization." The Always allow fix is in the Fixes list.

  9. Conductor 0.89.0: Sign in with ChatGPT

    Finding: Sign in with ChatGPT for Codex; "Steer Cursor chats while they're running, in local and cloud workspaces."; PR view top card; merge blocked if the PR changed on GitHub; Codex tasks no longer stop on temporary rate-limit errors; custom OAuth MCP servers can use the Conductor callback URL. PR view: "open the Review row to add unresolved comments to chat, or the Checks row to re-run or fix failed checks."

  10. Conductor.build on the App Store

    Finding: US App Store lists Conductor.build by Melty, Inc., iPhone only, free, requires iOS 26.0+. Current displayed version 1.1.3 is labelled 2d ago on November 7; no exact release date is inferred. History explicitly lists 1.0.5 October 8 and 1.1.0 October 23, with no year displayed. Store listing does not override the sign-up closure.

conductor:COORD-06: Home filters and sidebar sections organize workspaces; a matrix-style status indicator was introduced as experimental opt-in (0.85.0), and 0.90.0 says cloud organizations show it in Home, the sidebar and workspace tabs. Fleet usability was not tested. cross-checked

conductor:COORD-06

Home filters and sidebar sections organize workspaces; a matrix-style status indicator was introduced as experimental opt-in (0.85.0), and 0.90.0 says cloud organizations show it in Home, the sidebar and workspace tabs. Fleet usability was not tested.

  • Experimental status matrix; no fleet-scale test.
  1. Multiplayer

    Finding: Cloud organization workspaces are shared with the team; transcripts update live; presence dots for active teammates and typing indicators in a shared chat; Home filters by teammate and project; follow and reassign workspaces; direct links. The page does not describe co-prompting, private workspaces, roles or plan limits. Page undated.

  2. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  3. Conductor 0.90.0: Conductor for iOS

    Finding: iOS app (App Store link id6791228564): chat with agents and merge PRs from a phone, works with cloud workspaces. In cloud organizations pinned workspaces sync across devices and Automerge, optional-check and PR-order settings carry over to iOS. Notifications show the chat title instead of the branch name (no client named). Misc: "Pausing, resuming, or clearing a Codex goal shows up in the chat." Undated "New in 0.90.1" notes: background-command chats show as waiting and queued messages still send. No plan or iOS version requirement stated.

conductor:COORD-07: Per-session harness, model, Fast Mode and reasoning-level controls exist; workflow presets, coordinator and child-agent roles and natural-language orchestration are not documented. cross-checked

conductor:COORD-07

Per-session harness, model, Fast Mode and reasoning-level controls exist; workflow presets, coordinator and child-agent roles and natural-language orchestration are not documented.

  • Controls apply per chat tab; no coordinator or child-agent configuration was located.
  • Cloud Computer natural-language setup configures the environment, not agent workflows.
  1. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

  2. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

  3. Agent behavior

    Finding: Session controls table: Plan Mode, Fast Mode ("where supported"), thinking or reasoning level, Codex personality, checkpoints; no per-agent support listed. Repository preferences for review, PR, error-fix, conflict resolution and branch rename; agents read AGENTS.md, CLAUDE.md, .claude/commands and skills.

  4. Conductor 0.84.0: Edit files directly in diffs

    Finding: Edit files in diffs; effort change mid-chat for some Claude models; "Codex chats now retry automatically when OpenAI rate-limits a request" (no surface named); "Port forwarding is now off by default for each workspace"; PR links can open in GitHub or Linear (Settings → Git); 0.84.1 swaps cloud agent auth without restart; remote desktop control in cloud (experimental).

conductor:RUN-01: Managed Cloud (Vercel sandbox, us-east-1) keeps agents running after the Mac app closes until sandbox sleep or expiry; local sessions end when the app closes or the Mac shuts down; own and self-hosted cloud machines are not available. Cloud workspace disk is 32 GB; the Cloud Computer disk is 64 GB per one dated changelog entry. cross-checked

conductor:RUN-01

Managed Cloud (Vercel sandbox, us-east-1) keeps agents running after the Mac app closes until sandbox sleep or expiry; local sessions end when the app closes or the Mac shuts down; own and self-hosted cloud machines are not available. Cloud workspace disk is 32 GB; the Cloud Computer disk is 64 GB per one dated changelog entry.

  • Sandbox sleeps after 20 minutes without agent or terminal activity and stops at 23h50m.
  • Disk sizes are scoped: cloud workspaces 32 GB (undated docs); Cloud Computer 64 GB from one dated changelog entry that the docs do not repeat. No hands-on check of either disk.
  • Behavior across a disconnect was not tested.
  • Continuation from another device is documented but was not tested.
  1. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  2. Conductor Cloud

    Finding: Each agent runs in its own isolated cloud sandbox and keeps working after the laptop closes until the sandbox sleeps; "The chat, diff, git, terminal, and pull request workflow is the same in both" local and cloud. Support via Help > Send Feedback or humans@conductor.build. Page undated. Local vs Cloud table: "After your Mac closes | Local processes stop | Agents keep running until the sandbox sleeps".

  3. Cloud workspaces

    Finding: Cloud inactivity sleep is now 20 minutes without agent or terminal activity; focused window counts as activity. Maximum lifetime remains 23h50m even when active. Files and chats survive sleep; processes stop and interrupted turns must be resent.

  4. Conductor 0.90.0: Conductor for iOS

    Finding: iOS app (App Store link id6791228564): chat with agents and merge PRs from a phone, works with cloud workspaces. In cloud organizations pinned workspaces sync across devices and Automerge, optional-check and PR-order settings carry over to iOS. Notifications show the chat title instead of the branch name (no client named). Misc: "Pausing, resuming, or clearing a Codex goal shows up in the chat." Undated "New in 0.90.1" notes: background-command chats show as waiting and queued messages still send. No plan or iOS version requirement stated.

  5. Harnesses

    Finding: Four harnesses: Claude Code, Codex, Cursor, OpenCode. Claude Code, Codex and OpenCode are bundled; Cursor sessions use Cursor Agent through the Cursor API. One harness and model per chat tab. "Conductor is free to use. We do not bill or resell model usage." Page undated.

  6. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  7. Cloud Computer

    Finding: Each cloud workspace runs in an isolated microVM with 8 vCPU, 16 GB RAM and 32 GB ephemeral NVMe. The Cloud Computer is the shared development environment for a Conductor organization; the page gives no Cloud Computer disk size; up to 50 GitHub repositories; successful builds seed new workspaces; saved secrets are hidden but exposed as env vars inside workspaces; an Admin workspace agent can edit install/setup scripts and run builds.

  8. Conductor 0.82.0: Conductor MCP

    Finding: Conductor MCP launched; cloud MCP servers; "Your Cloud Computer got more oomph - they now have double the disk space (64GB)"; low-disk warnings.

  9. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  10. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

  11. Conductor pricing redirect

    Finding: The pricing URL redirects to the homepage, which advertises Cloud, desktop and mobile and links Get Conductor. No plan price table appears. This does not establish signup completion, upgrade eligibility or free-tier entitlements.

  12. Conductor Cloud FAQ

    Finding: Cloud access is described for Pro, Teams and Enterprise; Team signup is invite-only. The FAQ states no additional Cloud compute fee currently and plans usage-based compute pricing. Users bring subscriptions or keys. It also describes managed Vercel sandboxes and a planned self-hosted offering, without release dates. The mobile coming-soon wording is stale against the released iOS app.

  13. Configure model providers

    Finding: Model usage is billed through the selected provider account, subscription, usage pool or key. Conductor does not bill or resell it.

  14. What is Conductor Cloud?

    Finding: The Mac app supports local and Cloud workspaces. Local processes stop when the Mac closes; Cloud agents continue until sandbox sleep. The page still documents shared Cloud workspaces and setup.

  15. Conductor Multiplayer

    Finding: Cloud organization members share workspaces and chats, see presence and live output, follow workspaces, reassign work and open direct links. The current page does not impose the withdrawn shutdown restriction.

conductor:RUN-02: Local workspaces use a git worktree and branch each without an OS sandbox; Cloud uses isolated Linux microVM sandboxes. cross-checked

conductor:RUN-02

Local workspaces use a git worktree and branch each without an OS sandbox; Cloud uses isolated Linux microVM sandboxes.

  • Network isolation and container use not documented.
  • Isolation is described as development isolation, not a security boundary.
  • Security page says enabling permissions lets cloud agents act on the local Mac.
  1. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  2. First workspace

    Finding: Repository options: "Open project" (existing local repository), "Open GitHub project" and "Quick start" (new repository). Setup and run scripts; chats with Claude Code, Codex, Cursor or OpenCode; gh auth status check. No text on repositories without a remote or non-git folders. Page undated.

  3. Security and permissions

    Finding: Cloud agents run in isolated Linux sandboxes hosted by Vercel and can use the workspace environment variables and secrets; repository files and session data are kept in Conductor-managed infrastructure. Local agents run with the user account permissions; approvals can be required for shell commands, file changes, MCP tool calls or web fetches. Page undated.

  4. Isolated workspaces

    Finding: Each workspace is a git worktree on its own branch; a branch can be checked out in one workspace at a time; states workspace isolation is development isolation, not a security boundary, and agents run on the Mac with user permissions.

  5. Cloud Computer

    Finding: Each cloud workspace runs in an isolated microVM with 8 vCPU, 16 GB RAM and 32 GB ephemeral NVMe. The Cloud Computer is the shared development environment for a Conductor organization; the page gives no Cloud Computer disk size; up to 50 GitHub repositories; successful builds seed new workspaces; saved secrets are hidden but exposed as env vars inside workspaces; an Admin workspace agent can edit install/setup scripts and run builds.

  6. FAQ

    Finding: Agents run unsandboxed with the user account access; auth reuses Claude Code/Codex logins with an API-key mode. "How does Conductor make money? Right now we don't." and "At some point we plan to charge for collaboration features". Moving computers: "Conductor stores files at ~/Library/Application Support/com.conductor.app and ~/conductor." Page undated.

  7. Testing

    Finding: Run scripts use CONDUCTOR_PORT so workspaces run side by side; Spotlight testing syncs one workspace back to the repository root; automation covers one service at a time.

  8. Conductor 0.78.0: Introducing Conductor Cloud

    Finding: Cloud workspaces on isolated microVMs that keep running when the laptop closes; start by signing up for Pro and creating an organization; shareable links, presence and teammates prompting agents together in real time; API to manage cloud workspaces. Said the mobile app was "coming soon" on 2026-07-30. "To get started, sign up for Pro and create an organization." Billing "opens the Stripe billing portal". No price stated.

conductor:RUN-03: Cloud preserves files and chat history across sleep but not running processes; interrupted turns must be resent. cross-checked

conductor:RUN-03

Cloud preserves files and chat history across sleep but not running processes; interrupted turns must be resent.

  • Crash and restart behavior not tested.
  1. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  2. Cloud workspaces

    Finding: Cloud inactivity sleep is now 20 minutes without agent or terminal activity; focused window counts as activity. Maximum lifetime remains 23h50m even when active. Files and chats survive sleep; processes stop and interrupted turns must be resent.

  3. Conductor 0.83.0: Auto port forwarding

    Finding: Cloud auto port forwarding; one-click PR approval from the git panel; messages to sleeping cloud workspaces wake them; View deployment(s) button; added completion sounds; "Fixed notifications for workspaces in another organization opening the dashboard instead of the chat" (no client named).

  4. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  5. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

conductor:RUN-04: Cloud Computer builds seed new environments; setup scripts run per workspace; port forwarding is off by default for each workspace. cross-checked

conductor:RUN-04

Cloud Computer builds seed new environments; setup scripts run per workspace; port forwarding is off by default for each workspace.

  • Cache, service and repair behavior not documented.
  • Environment operations differ: 0.80.0 says editing cloud environment variables needs no rebuild; Cloud Computer docs specify deleting a saved row takes effect in the next build, and builds seed new workspaces. These statements do not establish one universal update rule; not runtime-tested.
  1. Conductor Cloud

    Finding: Each agent runs in its own isolated cloud sandbox and keeps working after the laptop closes until the sandbox sleeps; "The chat, diff, git, terminal, and pull request workflow is the same in both" local and cloud. Support via Help > Send Feedback or humans@conductor.build. Page undated. Local vs Cloud table: "After your Mac closes | Local processes stop | Agents keep running until the sandbox sleeps".

  2. Cloud workspaces

    Finding: Four-hour inactivity sleep; 23h50m hard lifetime. Files/chat persist, processes stop and interrupted turns need resending.

  3. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  4. Cloud Computer

    Finding: Each cloud workspace runs in an isolated microVM with 8 vCPU, 16 GB RAM and 32 GB ephemeral NVMe. The Cloud Computer is the shared development environment for a Conductor organization; the page gives no Cloud Computer disk size; up to 50 GitHub repositories; successful builds seed new workspaces; saved secrets are hidden but exposed as env vars inside workspaces; an Admin workspace agent can edit install/setup scripts and run builds.

  5. Set up cloud

    Finding: Cloud onboarding: create organization, connect the Conductor GitHub App, add an agent credential (subscription token or API key), build the Cloud Computer, then work in an Admin workspace. Admins can add shared API keys that take precedence over personal keys.

  6. Testing

    Finding: Run scripts use CONDUCTOR_PORT so workspaces run side by side; Spotlight testing syncs one workspace back to the repository root; automation covers one service at a time.

  7. Conductor 0.80.0: Stacks

    Finding: GitHub stacks (cloud; local needs gh stack extension); cloud workspaces include the Conductor CLI and Chrome; cloud environment variables moved to Organization → Environment and "editing them no longer requires rebuilding your Cloud Computer"; GH_HOST respected for GitHub Enterprise.

  8. Conductor 0.83.0: Auto port forwarding

    Finding: Cloud auto port forwarding; one-click PR approval from the git panel; messages to sleeping cloud workspaces wake them; View deployment(s) button; added completion sounds; "Fixed notifications for workspaces in another organization opening the dashboard instead of the chat" (no client named).

  9. Conductor 0.84.0: Edit files directly in diffs

    Finding: Edit files in diffs; effort change mid-chat for some Claude models; "Codex chats now retry automatically when OpenAI rate-limits a request" (no surface named); "Port forwarding is now off by default for each workspace"; PR links can open in GitHub or Linear (Settings → Git); 0.84.1 swaps cloud agent auth without restart; remote desktop control in cloud (experimental).

conductor:RUN-05: Cloud sandboxes enforce sleep and lifetime limits; Codex chats retry on OpenAI rate limits (0.84.0, no surface named). cross-checked

conductor:RUN-05

Cloud sandboxes enforce sleep and lifetime limits; Codex chats retry on OpenAI rate limits (0.84.0, no surface named).

  • Turn, token and spend limits not documented.
  1. Cloud workspaces

    Finding: Cloud inactivity sleep is now 20 minutes without agent or terminal activity; focused window counts as activity. Maximum lifetime remains 23h50m even when active. Files and chats survive sleep; processes stop and interrupted turns must be resent.

  2. Conductor 0.84.0: Edit files directly in diffs

    Finding: Edit files in diffs; effort change mid-chat for some Claude models; "Codex chats now retry automatically when OpenAI rate-limits a request" (no surface named); "Port forwarding is now off by default for each workspace"; PR links can open in GitHub or Linear (Settings → Git); 0.84.1 swaps cloud agent auth without restart; remote desktop control in cloud (experimental).

  3. Conductor 0.89.0: Sign in with ChatGPT

    Finding: Sign in with ChatGPT for Codex; "Steer Cursor chats while they're running, in local and cloud workspaces."; PR view top card; merge blocked if the PR changed on GitHub; Codex tasks no longer stop on temporary rate-limit errors; custom OAuth MCP servers can use the Conductor callback URL. PR view: "open the Review row to add unresolved comments to chat, or the Checks row to re-run or fix failed checks."

  4. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  5. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

conductor:RUN-06: API session status distinguishes idle, working and errored; failed Cloud startup exposes setup logs. API errors include userMessage, and release 0.84.0 shows GitHub's failure reason when an Actions job does not start. Traces and usage statistics remain unestablished. cross-checked

conductor:RUN-06

API session status distinguishes idle, working and errored; failed Cloud startup exposes setup logs. API errors include userMessage, and release 0.84.0 shows GitHub's failure reason when an Actions job does not start. Traces and usage statistics remain unestablished.

  • Status states were not checked against agent behaviour under an induced failure or approval wait.
  • No traces, usage exports or operational statistics are documented.
  1. Conductor API

    Finding: Beta HTTP API under https://api.conductor.build/v0 with an OpenAPI document; bearer API key; endpoints for projects, workspaces (create, get, rename, archive, unarchive, sleep), sessions, prompts, transcripts, cancel ("drop queued messages"), SQL and routines. Routines created through the API run a saved prompt in a fresh workspace on each webhook JSON POST. Cookbook: "messages sent while it works are steered into the running turn"; Tips: a queued prompt reports idle until its turn starts. Some 403s come from a proxy that rejects default client signatures (send a User-Agent). Shows a routine example with the `conductor` CLI. No SDK is documented on this page. Quick start (for AI) requires a Conductor Pro account to generate an API key. Page undated. Endpoint table: GET /v0/sessions/{id}/status "Whether the agent is idle, working, or errored"; POST /v0/sessions/{id}/cancel "Stop the agent's current turn and drop queued messages". Tips: "Always pass an explicit model" (workspace creation, POST /v0/workspaces) and "An optional effort field sets the thinking level for claude and codex sessions". No sidebar-section endpoint listed.

  2. Conductor MCP server

    Finding: Beta hosted MCP ("Tools and input shapes may change"): OAuth limited to mcp:tools and one organization, or bearer API key. Tools include create_session, send_message ("Send or queue a prompt for a session"), cancel_session ("Cancel the current agent turn and discard queued messages"), get_session_status ("Get a session's current agent status", no values listed), list_models, list_messages (after cursor), get_message and run_sql. No routine or section tools listed. Page undated.

  3. Conductor 0.77.0

    Finding: Alpha multiplayer and API for Pro users; "Workspace notifications now show the workspace name instead of the branch name" (no client named); experimental cloud cost estimates include Codex usage; background tasks live for everyone. API alpha: "Available to all Pro users." No price stated.

  4. Conductor 0.84.0: Edit files directly in diffs

    Finding: Edit files in diffs; effort change mid-chat for some Claude models; "Codex chats now retry automatically when OpenAI rate-limits a request" (no surface named); "Port forwarding is now off by default for each workspace"; PR links can open in GitHub or Linear (Settings → Git); 0.84.1 swaps cloud agent auth without restart; remote desktop control in cloud (experimental).

  5. Conductor 0.87.0: Search across files, OpenCode 2.0

    Finding: Routine GitHub trigger filters (label, author, base branch) and weekday schedules; OpenCode 2.0; "Create and delete sidebar sections from the CLI, API, and MCP in cloud organizations"; with the experimental status matrix enabled, indicators "show distinct states for routine runs, attention, errors, running pull request checks, and merged pull requests"; video thumbnails in chat; security fixes; setup scripts from shared cloud workspaces no longer auto-run on collaborators' Macs.

conductor:RUN-07: Experimental cloud cost estimates (Codex, 0.77.0) exist; per-task cost reporting is otherwise not documented; 0.86.0: "The context usage popover shows Claude account limits and reset times, matching Codex." cross-checked

conductor:RUN-07

Experimental cloud cost estimates (Codex, 0.77.0) exist; per-task cost reporting is otherwise not documented; 0.86.0: "The context usage popover shows Claude account limits and reset times, matching Codex."

  • Only a changelog mention; scope and accuracy not verified.
  1. Conductor 0.77.0

    Finding: Alpha multiplayer and API for Pro users; "Workspace notifications now show the workspace name instead of the branch name" (no client named); experimental cloud cost estimates include Codex usage; background tasks live for everyone. API alpha: "Available to all Pro users." No price stated.

  2. Conductor 0.86.0: Share your loadouts

    Finding: "Removed Codex goals and message queuing. Messages sent while an agent is working now steer it instead of waiting in a queue" (no surface named); answer Codex questions while it works (no surface named); context usage popover shows Claude account limits and reset times, matching Codex; cloud run scripts; private workspaces marked with a lock; fixes for OpenCode rejecting Always allow approvals and for "cloud plan approvals appearing successful when the workspace rejected the response". Improvements also: "Open pull requests in Graphite"; "Prompt links fill the new-workspace composer in your active organization." The Always allow fix is in the Fixes list.

conductor:RUN-08: Each Cloud workspace is documented with 8 vCPUs and 16 GB RAM; concurrency limits, resource-aware scheduling and fairness guarantees remain unestablished. cross-checked

conductor:RUN-08

Each Cloud workspace is documented with 8 vCPUs and 16 GB RAM; concurrency limits, resource-aware scheduling and fairness guarantees remain unestablished.

  • No concurrency, backpressure or fairness scenario was run.
  1. Cloud Computer

    Finding: Each cloud workspace runs in an isolated microVM with 8 vCPU, 16 GB RAM and 32 GB ephemeral NVMe. The Cloud Computer is the shared development environment for a Conductor organization; the page gives no Cloud Computer disk size; up to 50 GitHub repositories; successful builds seed new workspaces; saved secrets are hidden but exposed as env vars inside workspaces; an Admin workspace agent can edit install/setup scripts and run builds.

conductor:TEAM-01: Shared Cloud workspaces support live chats, presence and typing indicators; private workspaces exist; Pro collaboration covers up to five Pro users. cross-checked

conductor:TEAM-01

Shared Cloud workspaces support live chats, presence and typing indicators; private workspaces exist; Pro collaboration covers up to five Pro users.

  • Input ordering and conflicts were not tested.
  1. Pricing

    Finding: Plans: Free $0 (local workspaces on a Mac, own subscriptions and keys); Pro $50/mo with no per-user or per-account qualifier (Conductor Cloud and "a bunch of cloud workspace hours", multiplayer with up to 5 Pro users, Conductor API, "Conductor mobile app"); Teams $60/mo/user, invite-only (admin portal, centralized billing); Enterprise custom (DPA, SAML SSO, SCIM, SLA). No trial or card text. FAQ: no extra Cloud fee now, usage-based compute pricing planned. SOC 2 Type II statement. Page undated.

  2. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  3. Multiplayer

    Finding: Cloud organization workspaces are shared with the team; transcripts update live; presence dots for active teammates and typing indicators in a shared chat; Home filters by teammate and project; follow and reassign workspaces; direct links. The page does not describe co-prompting, private workspaces, roles or plan limits. Page undated.

  4. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  5. Conductor 0.78.0: Introducing Conductor Cloud

    Finding: Cloud workspaces on isolated microVMs that keep running when the laptop closes; start by signing up for Pro and creating an organization; shareable links, presence and teammates prompting agents together in real time; API to manage cloud workspaces. Said the mobile app was "coming soon" on 2026-07-30. "To get started, sign up for Pro and create an organization." Billing "opens the Stripe billing portal". No price stated.

conductor:TEAM-02: Cloud organizations have admins and members. Organization admins can configure shared agent API keys, while members provide personal credentials. Current SSO protocol and enterprise provisioning details are not reestablished by this pass.

conductor:TEAM-02

Cloud organizations have admins and members. Organization admins can configure shared agent API keys, while members provide personal credentials. Current SSO protocol and enterprise provisioning details are not reestablished by this pass.

  • First-party documentation review only; no multi-account permission or enterprise-login scenario was run. Retained task-role findings keep their older cited scope.
  1. Pricing

    Finding: Plans: Free $0 (local workspaces on a Mac, own subscriptions and keys); Pro $50/mo with no per-user or per-account qualifier (Conductor Cloud and "a bunch of cloud workspace hours", multiplayer with up to 5 Pro users, Conductor API, "Conductor mobile app"); Teams $60/mo/user, invite-only (admin portal, centralized billing); Enterprise custom (DPA, SAML SSO, SCIM, SLA). No trial or card text. FAQ: no extra Cloud fee now, usage-based compute pricing planned. SOC 2 Type II statement. Page undated.

  2. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  3. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  4. Set up cloud

    Finding: Cloud onboarding: create organization, connect the Conductor GitHub App, add an agent credential (subscription token or API key), build the Cloud Computer, then work in an Admin workspace. Admins can add shared API keys that take precedence over personal keys.

  5. Conductor Cloud runtime FAQ

    Finding: Conductor provisions a Vercel sandbox for each Cloud workspace, keeps Cloud sessions on its servers, and does not yet support user-supplied cloud machines. Local sessions stop when the Mac closes. The plan-family section includes an outdated mobile-app coming-soon sentence, so this pass does not use it to reconstruct current commercial entitlements. The Enterprise description names automated user provisioning without specifying its protocol.

  6. Conductor Cloud organization onboarding

    Finding: Members joining an already configured organization add their agent credentials as needed. Organization admins can additionally configure shared agent API keys for the organization; those keys override a member personal key for the same agent. Organization creation includes optional teammate invitations.

conductor:TEAM-03: Local tool approvals exist, with settings for defaults, plan mode and managed enforcement. cross-checked

conductor:TEAM-03

Local tool approvals exist, with settings for defaults, plan mode and managed enforcement.

  • Network and secret approvals not documented.
  1. Security and permissions

    Finding: Cloud agents run in isolated Linux sandboxes hosted by Vercel and can use the workspace environment variables and secrets; repository files and session data are kept in Conductor-managed infrastructure. Local agents run with the user account permissions; approvals can be required for shell commands, file changes, MCP tool calls or web fetches. Page undated.

  2. Agent modes

    Finding: Model picker for Claude Code, Codex, Cursor or OpenCode (Cursor sessions support Composer 2.5); controls can change "before or during a chat". Plan Mode: plan before editing, approve or give feedback; the Supported controls table lists Plan Mode and checkpoints for Claude Code and Codex, and marks both unsupported for Cursor and OpenCode. Codex goals: pause, resume, edit or clear via the goal bar, "available for Codex sessions in local workspaces", "Cloud workspaces do not support goals yet". Skills: Codex, Claude Code and OpenCode. Page undated.

  3. Managed settings

    Finding: Provisional organization-controlled settings file on a Mac overrides user and repository settings (for example enterprise_data_privacy, default models, environment variables).

  4. Settings reference

    Finding: Settings include tool_approvals_enabled, Codex tool_approval_policy (default or always_approve), default plan mode, review/PR/conflict prompts, git.archive_on_merge, enterprise_data_privacy.

  5. Conductor 0.86.0: Share your loadouts

    Finding: "Removed Codex goals and message queuing. Messages sent while an agent is working now steer it instead of waiting in a queue" (no surface named); answer Codex questions while it works (no surface named); context usage popover shows Claude account limits and reset times, matching Codex; cloud run scripts; private workspaces marked with a lock; fixes for OpenCode rejecting Always allow approvals and for "cloud plan approvals appearing successful when the workspace rejected the response". Improvements also: "Open pull requests in Graphite"; "Prompt links fill the new-workspace composer in your active organization." The Always allow fix is in the Fixes list.

conductor:TEAM-04: The revised announcement documents transferring Cloud workspaces to local and exporting an account archive. The historical 0.93.0 notes explicitly include chats in the transfer. Audit integrity and export format were not established. cross-checked

conductor:TEAM-04

The revised announcement documents transferring Cloud workspaces to local and exporting an account archive. The historical 0.93.0 notes explicitly include chats in the transfer. Audit integrity and export format were not established.

  • Export format, completeness and integrity guarantees are not established; no export was executed.
  1. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  2. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

  3. Conductor 0.93.0 historical transition

    Finding: The November 4 release still lists Cloud read-only November 13, shutdown November 20 and local model/fix updates until early 2027. These historical plans are superseded by the revised announcement and 0.94.2 banner removal. It also documents transferring a workspace and chats to local.

conductor:TEAM-05: Diff viewer, line comments, GitHub review threads and suggested PR actions guide review. cross-checked

conductor:TEAM-05

Diff viewer, line comments, GitHub review threads and suggested PR actions guide review.

  • Request-changes and batch review not documented.
  1. Workflow

    Finding: Workspace is the delegation unit; branch/PR is integration unit. GitHub/Linear intake and repository instructions are documented.

  2. Diff viewer

    Finding: Unified/commit-filtered diffs, line feedback and GitHub review comments. "Conductor recommends actions as a workspace moves toward merge. Use those actions to create a pull request, respond to feedback, fix checks, and merge". Page undated.

  3. Checks

    Finding: Checks may include Git status, "Pull request metadata", CI and status checks, "Deployments", GitHub comments and review threads, and todos. Blockers: "Conductor may block or discourage merge actions when required work is still open, such as unresolved todos or failed checks", so enforcement strength is not established. Page undated.

  4. Review and merge a workspace

    Finding: Diff viewer, Review action (agent reviews diff), Checks tab; unresolved comments, failing checks and open todos are to be treated as blockers; merge, then archive.

conductor:TEAM-06: Teammates can continue and reassign a Cloud workspace. cross-checked

conductor:TEAM-06

Teammates can continue and reassign a Cloud workspace.

  • Handoff context not tested.
  1. Multiplayer

    Finding: Cloud organization workspaces are shared with the team; transcripts update live; presence dots for active teammates and typing indicators in a shared chat; Home filters by teammate and project; follow and reassign workspaces; direct links. The page does not describe co-prompting, private workspaces, roles or plan limits. Page undated.

  2. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

conductor:SHIP-01: Cloud workspaces show diffs; plans, screenshots and attachments in .context sync from cloud workspaces to the Mac (0.85.0); chat videos show a thumbnail that opens the media viewer (0.87.0). cross-checked

conductor:SHIP-01

Cloud workspaces show diffs; plans, screenshots and attachments in .context sync from cloud workspaces to the Mac (0.85.0); chat videos show a thumbnail that opens the media viewer (0.87.0).

  • Recordings, test outputs not documented in detail.
  1. Multiplayer

    Finding: Cloud organization workspaces are shared with the team; transcripts update live; presence dots for active teammates and typing indicators in a shared chat; Home filters by teammate and project; follow and reassign workspaces; direct links. The page does not describe co-prompting, private workspaces, roles or plan limits. Page undated.

  2. Cloud workspaces

    Finding: Four-hour inactivity sleep; 23h50m hard lifetime. Files/chat persist, processes stop and interrupted turns need resending.

  3. Diff viewer

    Finding: Unified/commit-filtered diffs, line feedback and GitHub review comments. "Conductor recommends actions as a workspace moves toward merge. Use those actions to create a pull request, respond to feedback, fix checks, and merge". Page undated.

  4. Conductor 0.85.0

    Finding: Routines run agents on a schedule or via GitHub action, in the cloud, so Pro and Team only (no webhook text on this page); private workspaces; favorite models sync across devices in cloud organizations; Cursor subscription sign-in for cloud workspaces; per-organization cloud agent connections; "Plans, screenshots, and attachments in .context now sync from cloud workspaces to your Mac"; experimental workspace status matrix. Model picker shortcuts change models, effort and speed.

  5. Conductor 0.90.0: Conductor for iOS

    Finding: iOS app (App Store link id6791228564): chat with agents and merge PRs from a phone, works with cloud workspaces. In cloud organizations pinned workspaces sync across devices and Automerge, optional-check and PR-order settings carry over to iOS. Notifications show the chat title instead of the branch name (no client named). Misc: "Pausing, resuming, or clearing a Codex goal shows up in the chat." Undated "New in 0.90.1" notes: background-command chats show as waiting and queued messages still send. No plan or iOS version requirement stated.

  6. Privacy

    Finding: Cloud session inputs and outputs are stored on Conductor servers; local session data stays on the Mac (~/Library/Application Support/com.conductor.app); analytics and crash logs via PostHog, including OS and IP address. No export, deletion or retention text. Enterprise data privacy is local-only and "cannot be used in cloud". Page undated.

  7. Conductor 0.83.0: Auto port forwarding

    Finding: Cloud auto port forwarding; one-click PR approval from the git panel; messages to sleeping cloud workspaces wake them; View deployment(s) button; added completion sounds; "Fixed notifications for workspaces in another organization opening the dashboard instead of the chat" (no client named).

  8. Conductor 0.84.0: Edit files directly in diffs

    Finding: Edit files in diffs; effort change mid-chat for some Claude models; "Codex chats now retry automatically when OpenAI rate-limits a request" (no surface named); "Port forwarding is now off by default for each workspace"; PR links can open in GitHub or Linear (Settings → Git); 0.84.1 swaps cloud agent auth without restart; remote desktop control in cloud (experimental).

  9. Conductor 0.87.0: Search across files, OpenCode 2.0

    Finding: Routine GitHub trigger filters (label, author, base branch) and weekday schedules; OpenCode 2.0; "Create and delete sidebar sections from the CLI, API, and MCP in cloud organizations"; with the experimental status matrix enabled, indicators "show distinct states for routine runs, attention, errors, running pull request checks, and merged pull requests"; video thumbnails in chat; security fixes; setup scripts from shared cloud workspaces no longer auto-run on collaborators' Macs.

conductor:SHIP-02: Checks aggregate CI; unresolved todos block merge; the checks page says Conductor "may block or discourage" merge for failed checks; 0.89.0 blocks a merge when the PR on GitHub has changed. cross-checked

conductor:SHIP-02

Checks aggregate CI; unresolved todos block merge; the checks page says Conductor "may block or discourage" merge for failed checks; 0.89.0 blocks a merge when the PR on GitHub has changed.

  • How blocking is decided is not documented; failed-check blocking is not established.
  1. Checks

    Finding: Checks may include Git status, "Pull request metadata", CI and status checks, "Deployments", GitHub comments and review threads, and todos. Blockers: "Conductor may block or discourage merge actions when required work is still open, such as unresolved todos or failed checks", so enforcement strength is not established. Page undated.

  2. Todos

    Finding: Humans and agents use workspace todos; unchecked todos block merge; @todos exposes list to agent.

  3. Conductor 0.90.0: Conductor for iOS

    Finding: iOS app (App Store link id6791228564): chat with agents and merge PRs from a phone, works with cloud workspaces. In cloud organizations pinned workspaces sync across devices and Automerge, optional-check and PR-order settings carry over to iOS. Notifications show the chat title instead of the branch name (no client named). Misc: "Pausing, resuming, or clearing a Codex goal shows up in the chat." Undated "New in 0.90.1" notes: background-command chats show as waiting and queued messages still send. No plan or iOS version requirement stated.

  4. Review and merge a workspace

    Finding: Diff viewer, Review action (agent reviews diff), Checks tab; unresolved comments, failing checks and open todos are to be treated as blockers; merge, then archive.

  5. Testing

    Finding: Run scripts use CONDUCTOR_PORT so workspaces run side by side; Spotlight testing syncs one workspace back to the repository root; automation covers one service at a time.

  6. Conductor 0.89.0: Sign in with ChatGPT

    Finding: Sign in with ChatGPT for Codex; "Steer Cursor chats while they're running, in local and cloud workspaces."; PR view top card; merge blocked if the PR changed on GitHub; Codex tasks no longer stop on temporary rate-limit errors; custom OAuth MCP servers can use the Conductor callback URL. PR view: "open the Review row to add unresolved comments to chat, or the Checks row to re-run or fix failed checks."

conductor:SHIP-03: Workspace branch-to-PR workflow is documented, with GitHub stacks (0.80.0; local needs the gh stack extension) and an option to open PRs in Graphite (0.86.0). cross-checked

conductor:SHIP-03

Workspace branch-to-PR workflow is documented, with GitHub stacks (0.80.0; local needs the gh stack extension) and an option to open PRs in Graphite (0.86.0).

  • Rebase and commit handling not documented.
  • The fix-checks action is documented without describing what it sends to the agent; not tested.
  1. First workspace

    Finding: Repository options: "Open project" (existing local repository), "Open GitHub project" and "Quick start" (new repository). Setup and run scripts; chats with Claude Code, Codex, Cursor or OpenCode; gh auth status check. No text on repositories without a remote or non-git folders. Page undated.

  2. Diff viewer

    Finding: Unified/commit-filtered diffs, line feedback and GitHub review comments. "Conductor recommends actions as a workspace moves toward merge. Use those actions to create a pull request, respond to feedback, fix checks, and merge". Page undated.

  3. Isolated workspaces

    Finding: Each workspace is a git worktree on its own branch; a branch can be checked out in one workspace at a time; states workspace isolation is development isolation, not a security boundary, and agents run on the Mac with user permissions.

  4. From issue to pull request

    Finding: A workspace can be started from a GitHub or Linear issue and inherits its title, description and context; Create PR sends the diff and repository context to the agent to draft the PR. Issue-PR linking is not described. After the PR exists, use Checks and Diff Viewer "to respond to CI, deployments, review comments, and todos".

  5. Agent behavior

    Finding: Session controls table: Plan Mode, Fast Mode ("where supported"), thinking or reasoning level, Codex personality, checkpoints; no per-agent support listed. Repository preferences for review, PR, error-fix, conflict resolution and branch rename; agents read AGENTS.md, CLAUDE.md, .claude/commands and skills.

  6. Conductor 0.80.0: Stacks

    Finding: GitHub stacks (cloud; local needs gh stack extension); cloud workspaces include the Conductor CLI and Chrome; cloud environment variables moved to Organization → Environment and "editing them no longer requires rebuilding your Cloud Computer"; GH_HOST respected for GitHub Enterprise.

  7. Conductor 0.86.0: Share your loadouts

    Finding: "Removed Codex goals and message queuing. Messages sent while an agent is working now steer it instead of waiting in a queue" (no surface named); answer Codex questions while it works (no surface named); context usage popover shows Claude account limits and reset times, matching Codex; cloud run scripts; private workspaces marked with a lock; fixes for OpenCode rejecting Always allow approvals and for "cloud plan approvals appearing successful when the workspace rejected the response". Improvements also: "Open pull requests in Graphite"; "Prompt links fill the new-workspace composer in your active organization." The Always allow fix is in the Fixes list.

  8. Conductor 0.89.0: Sign in with ChatGPT

    Finding: Sign in with ChatGPT for Codex; "Steer Cursor chats while they're running, in local and cloud workspaces."; PR view top card; merge blocked if the PR changed on GitHub; Codex tasks no longer stop on temporary rate-limit errors; custom OAuth MCP servers can use the Conductor callback URL. PR view: "open the Review row to add unresolved comments to chat, or the Checks row to re-run or fix failed checks."

conductor:SHIP-04: One-click PR approval and merge from Conductor; merge blocked if the PR changed on GitHub. cross-checked

conductor:SHIP-04

One-click PR approval and merge from Conductor; merge blocked if the PR changed on GitHub.

  • Release and deploy authority not documented.
  1. Review and merge a workspace

    Finding: Diff viewer, Review action (agent reviews diff), Checks tab; unresolved comments, failing checks and open todos are to be treated as blockers; merge, then archive.

  2. Conductor 0.83.0: Auto port forwarding

    Finding: Cloud auto port forwarding; one-click PR approval from the git panel; messages to sleeping cloud workspaces wake them; View deployment(s) button; added completion sounds; "Fixed notifications for workspaces in another organization opening the dashboard instead of the chat" (no client named).

  3. Conductor 0.89.0: Sign in with ChatGPT

    Finding: Sign in with ChatGPT for Codex; "Steer Cursor chats while they're running, in local and cloud workspaces."; PR view top card; merge blocked if the PR changed on GitHub; Codex tasks no longer stop on temporary rate-limit errors; custom OAuth MCP servers can use the Conductor callback URL. PR view: "open the Review row to add unresolved comments to chat, or the Checks row to re-run or fix failed checks."

conductor:SHIP-05: Local checkpoints support destructive turn reversion; the Checks tab may include pull request metadata and deployments. cross-checked

conductor:SHIP-05

Local checkpoints support destructive turn reversion; the Checks tab may include pull request metadata and deployments.

  • Cloud checkpoint scope and retention not documented.
  1. Checks

    Finding: Checks may include Git status, "Pull request metadata", CI and status checks, "Deployments", GitHub comments and review threads, and todos. Blockers: "Conductor may block or discourage merge actions when required work is still open, such as unresolved todos or failed checks", so enforcement strength is not established. Page undated.

  2. Checkpoints

    Finding: Checkpoints are stored locally as private Git refs, captured "Before each supported agent responds" (agents not named); restoring a turn deletes that turn and later messages and reverts code changes since. Page undated.

  3. Conductor 0.83.0: Auto port forwarding

    Finding: Cloud auto port forwarding; one-click PR approval from the git panel; messages to sleeping cloud workspaces wake them; View deployment(s) button; added completion sounds; "Fixed notifications for workspaces in another organization opening the dashboard instead of the chat" (no client named).

  4. Conductor 0.84.0: Edit files directly in diffs

    Finding: Edit files in diffs; effort change mid-chat for some Claude models; "Codex chats now retry automatically when OpenAI rate-limits a request" (no surface named); "Port forwarding is now off by default for each workspace"; PR links can open in GitHub or Linear (Settings → Git); 0.84.1 swaps cloud agent auth without restart; remote desktop control in cloud (experimental).

  5. Conductor 0.89.0: Sign in with ChatGPT

    Finding: Sign in with ChatGPT for Codex; "Steer Cursor chats while they're running, in local and cloud workspaces."; PR view top card; merge blocked if the PR changed on GitHub; Codex tasks no longer stop on temporary rate-limit errors; custom OAuth MCP servers can use the Conductor callback URL. PR view: "open the Review row to add unresolved comments to chat, or the Checks row to re-run or fix failed checks."

conductor:OWN-01: Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

conductor:OWN-01

Conductor terms explicitly define the client-side and hosted Platform as proprietary software.

  • Terms establish the main product rights; third-party dependencies and user outputs retain their separate licenses.
  1. meltylabs/conductor-releases (GitHub)

    Finding: Public repository whose recursive tree (GitHub API, main, read 2026-10-09) contains only CHANGELOG.md; 208 commits; latest commit 2025-12-18; CHANGELOG.md newest entry 0.27.0 ("Dec 18th, 2025"), so it no longer tracks current releases; no releases, no description, no Discussions, no roadmap; issues enabled (29 open).

  2. Conductor terms

    Finding: The terms explicitly define the Platform as proprietary hosted AI automation software including client-side and hosted software. They retain vendor intellectual property and grant restricted platform access.

conductor:OWN-02: The local app keeps chats and files on the Mac (~/Library/Application Support/com.conductor.app and ~/conductor), which can be moved with Migration Assistant or manual copying; the revised announcement also documents Settings > Account > Export data. Archive format, import and account independence remain unestablished. cross-checked

conductor:OWN-02

The local app keeps chats and files on the Mac (~/Library/Application Support/com.conductor.app and ~/conductor), which can be moved with Migration Assistant or manual copying; the revised announcement also documents Settings > Account > Export data. Archive format, import and account independence remain unestablished.

  • Archive format, whether Export data includes Mac-only local chats, import behaviour and account independence are not established.
  1. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  2. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  3. Privacy

    Finding: Cloud session inputs and outputs are stored on Conductor servers; local session data stays on the Mac (~/Library/Application Support/com.conductor.app); analytics and crash logs via PostHog, including OS and IP address. No export, deletion or retention text. Enterprise data privacy is local-only and "cannot be used in cloud". Page undated.

  4. FAQ

    Finding: Agents run unsandboxed with the user account access; auth reuses Claude Code/Codex logins with an API-key mode. "How does Conductor make money? Right now we don't." and "At some point we plan to charge for collaboration features". Moving computers: "Conductor stores files at ~/Library/Application Support/com.conductor.app and ~/conductor." Page undated.

  5. Install

    Finding: macOS only (not available for Windows or Linux yet); setup checks GitHub auth and at least one of Claude, Codex or Cursor credentials.

  6. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  7. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

  8. Conductor pricing redirect

    Finding: The pricing URL redirects to the homepage, which advertises Cloud, desktop and mobile and links Get Conductor. No plan price table appears. This does not establish signup completion, upgrade eligibility or free-tier entitlements.

  9. Conductor FAQ

    Finding: The FAQ calls Conductor a free tool and describes future collaboration charges, without plan entitlements. That undated wording does not resolve the Cloud FAQ plan gates. Migration Assistant or copying the named local app and provider directories is documented for moving chats.

conductor:OWN-03: Local workspaces and chats stay on the Mac, model requests go to the configured provider, and analytics go to PostHog. The revised announcement retains the no-sale/no-training statement and no longer gives shutdown deletion dates. cross-checked

conductor:OWN-03

Local workspaces and chats stay on the Mac, model requests go to the configured provider, and analytics go to PostHog. The revised announcement retains the no-sale/no-training statement and no longer gives shutdown deletion dates.

  • A telemetry opt-out, consent flow, general retention period and deletion procedure were not established. Model-provider policies remain separate.
  1. Cloud FAQ

    Finding: Cloud on Pro, Teams and Enterprise; Teams invite-only with an admin portal "to manage user roles"; Teams members can collaborate "by sharing workspaces and prompting agents together in real time"; no extra fee now, usage-based compute pricing planned; sandboxes have 8 vCPUs, 16 GB RAM and 32 GB ephemeral NVMe; Vercel sandboxes in us-east-1; GitHub only; own cloud machines "Not yet", self-hosted offering being built. The Pro answer still says "our mobile app (coming soon)". Page undated.

  2. Security and permissions

    Finding: Cloud agents run in isolated Linux sandboxes hosted by Vercel and can use the workspace environment variables and secrets; repository files and session data are kept in Conductor-managed infrastructure. Local agents run with the user account permissions; approvals can be required for shell commands, file changes, MCP tool calls or web fetches. Page undated.

  3. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  4. Cloud Computer

    Finding: Each cloud workspace runs in an isolated microVM with 8 vCPU, 16 GB RAM and 32 GB ephemeral NVMe. The Cloud Computer is the shared development environment for a Conductor organization; the page gives no Cloud Computer disk size; up to 50 GitHub repositories; successful builds seed new workspaces; saved secrets are hidden but exposed as env vars inside workspaces; an Admin workspace agent can edit install/setup scripts and run builds.

  5. Privacy

    Finding: Cloud session inputs and outputs are stored on Conductor servers; local session data stays on the Mac (~/Library/Application Support/com.conductor.app); analytics and crash logs via PostHog, including OS and IP address. No export, deletion or retention text. Enterprise data privacy is local-only and "cannot be used in cloud". Page undated.

  6. Conductor privacy

    Finding: Local chats stay on the Mac; Cloud stores inputs, outputs and workspace state. App analytics include feature/model events and errors in PostHog. No telemetry opt-out is stated.

  7. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  8. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

  9. Conductor privacy

    Finding: Local chats remain on the Mac and model traffic goes to the provider. Conductor stores Cloud inputs and outputs; PostHog stores analytics. This reference does not provide a general retention period or a shutdown deletion schedule.

conductor:OWN-04: The pricing URL redirects to the homepage without a price table. Cloud FAQ describes plan gates and no additional compute fee for now; model charges stay with the provider. Current subscription prices and allowances remain unestablished. cross-checked

conductor:OWN-04

The pricing URL redirects to the homepage without a price table. Cloud FAQ describes plan gates and no additional compute fee for now; model charges stay with the provider. Current subscription prices and allowances remain unestablished.

  • Public download and plan documentation does not establish signup completion, current charges or upgrade eligibility. No authenticated billing test was performed.
  1. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  2. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  3. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

  4. Conductor pricing redirect

    Finding: The pricing URL redirects to the homepage, which advertises Cloud, desktop and mobile and links Get Conductor. No plan price table appears. This does not establish signup completion, upgrade eligibility or free-tier entitlements.

  5. Conductor Cloud FAQ

    Finding: Cloud access is described for Pro, Teams and Enterprise; Team signup is invite-only. The FAQ states no additional Cloud compute fee currently and plans usage-based compute pricing. Users bring subscriptions or keys. It also describes managed Vercel sandboxes and a planned self-hosted offering, without release dates. The mobile coming-soon wording is stale against the released iOS app.

  6. Configure model providers

    Finding: Model usage is billed through the selected provider account, subscription, usage pool or key. Conductor does not bill or resell it.

conductor:OWN-05: One-, two- and five-person scenarios cannot be priced from the checked public pages because current subscription prices, allowances and workloads are not established. Model costs remain separate. cross-checked

conductor:OWN-05

One-, two- and five-person scenarios cannot be priced from the checked public pages because current subscription prices, allowances and workloads are not established. Model costs remain separate.

  • Public download and plan documentation does not establish signup completion, current charges or upgrade eligibility. No authenticated billing test was performed.
  1. Configure model providers

    Finding: Claude Code via CLI auth or API key mode (ANTHROPIC_API_KEY) or compatible providers (OpenRouter, Vercel AI Gateway, Bedrock, GLM); Codex via "Codex sign-in, an OpenAI subscription, or an API key" (CODEX_API_KEY) or gateway; OpenCode via provider credentials; Cursor via Cursor API key. "Conductor does not bill or resell model usage." Local models and Vertex AI not covered. Page undated.

  2. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  3. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

  4. Conductor pricing redirect

    Finding: The pricing URL redirects to the homepage, which advertises Cloud, desktop and mobile and links Get Conductor. No plan price table appears. This does not establish signup completion, upgrade eligibility or free-tier entitlements.

  5. Conductor Cloud FAQ

    Finding: Cloud access is described for Pro, Teams and Enterprise; Team signup is invite-only. The FAQ states no additional Cloud compute fee currently and plans usage-based compute pricing. Users bring subscriptions or keys. It also describes managed Vercel sandboxes and a planned self-hosted offering, without release dates. The mobile coming-soon wording is stale against the released iOS app.

  6. Configure model providers

    Finding: Model usage is billed through the selected provider account, subscription, usage pool or key. Conductor does not bill or resell it.

conductor:OWN-06: Conductor has withdrawn its wind-down plan. The revised announcement says development and normal operation continue; 0.94.2 removes the Cloud shutdown banner. The dates in the older 0.93.0 notes are superseded. cross-checked

conductor:OWN-06

Conductor has withdrawn its wind-down plan. The revised announcement says development and normal operation continue; 0.94.2 removes the Cloud shutdown banner. The dates in the older 0.93.0 notes are superseded.

  • Public download and plan documentation does not establish signup completion, current charges or upgrade eligibility. No authenticated billing test was performed.
  1. Conductor

    Finding: Homepage describes Cloud workspaces, multiplayer, four harnesses and mobile/API access. Latest changelog card is 0.93.0 dated November 4, 2026, joining Meta; acquisition banner links the announcement.

  2. Conductor API

    Finding: Beta HTTP API under https://api.conductor.build/v0 with an OpenAPI document; bearer API key; endpoints for projects, workspaces (create, get, rename, archive, unarchive, sleep), sessions, prompts, transcripts, cancel ("drop queued messages"), SQL and routines. Routines created through the API run a saved prompt in a fresh workspace on each webhook JSON POST. Cookbook: "messages sent while it works are steered into the running turn"; Tips: a queued prompt reports idle until its turn starts. Some 403s come from a proxy that rejects default client signatures (send a User-Agent). Shows a routine example with the `conductor` CLI. No SDK is documented on this page. Quick start (for AI) requires a Conductor Pro account to generate an API key. Page undated. Endpoint table: GET /v0/sessions/{id}/status "Whether the agent is idle, working, or errored"; POST /v0/sessions/{id}/cancel "Stop the agent's current turn and drop queued messages". Tips: "Always pass an explicit model" (workspace creation, POST /v0/workspaces) and "An optional effort field sets the thinking level for claude and codex sessions". No sidebar-section endpoint listed.

  3. Conductor MCP server

    Finding: Beta hosted MCP ("Tools and input shapes may change"): OAuth limited to mcp:tools and one organization, or bearer API key. Tools include create_session, send_message ("Send or queue a prompt for a session"), cancel_session ("Cancel the current agent turn and discard queued messages"), get_session_status ("Get a session's current agent status", no values listed), list_models, list_messages (after cursor), get_message and run_sql. No routine or section tools listed. Page undated.

  4. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  5. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

  6. Conductor 0.93.0 historical transition

    Finding: The November 4 release still lists Cloud read-only November 13, shutdown November 20 and local model/fix updates until early 2027. These historical plans are superseded by the revised announcement and 0.94.2 banner removal. It also documents transferring a workspace and chats to local.

  7. Install Conductor

    Finding: The install guide provides Mac download/setup steps, explicitly excludes Windows/Linux, and requires GitHub and an agent provider. It does not state an existing-account-only restriction.

  8. Conductor pricing redirect

    Finding: The pricing URL redirects to the homepage, which advertises Cloud, desktop and mobile and links Get Conductor. No plan price table appears. This does not establish signup completion, upgrade eligibility or free-tier entitlements.

  9. Conductor Cloud FAQ

    Finding: Cloud access is described for Pro, Teams and Enterprise; Team signup is invite-only. The FAQ states no additional Cloud compute fee currently and plans usage-based compute pricing. Users bring subscriptions or keys. It also describes managed Vercel sandboxes and a planned self-hosted offering, without release dates. The mobile coming-soon wording is stale against the released iOS app.

  10. Configure model providers

    Finding: Model usage is billed through the selected provider account, subscription, usage pool or key. Conductor does not bill or resell it.

conductor:OWN-07: Conductor reports SOC 2 Type II; audit report and scope not inspected. cross-checked

conductor:OWN-07

Conductor reports SOC 2 Type II; audit report and scope not inspected.

  • Vendor assertion; scope not verified.
  1. Pricing

    Finding: Plans: Free $0 (local workspaces on a Mac, own subscriptions and keys); Pro $50/mo with no per-user or per-account qualifier (Conductor Cloud and "a bunch of cloud workspace hours", multiplayer with up to 5 Pro users, Conductor API, "Conductor mobile app"); Teams $60/mo/user, invite-only (admin portal, centralized billing); Enterprise custom (DPA, SAML SSO, SCIM, SLA). No trial or card text. FAQ: no extra Cloud fee now, usage-based compute pricing planned. SOC 2 Type II statement. Page undated.

conductor:OWN-09: Pricing is partially specified: model billing stays with the provider and Cloud FAQ states no additional compute fee currently. The pricing URL redirects without a table; subscription units, allowances, rates and cancellation terms remain unknown. cross-checked

conductor:OWN-09

Pricing is partially specified: model billing stays with the provider and Cloud FAQ states no additional compute fee currently. The pricing URL redirects without a table; subscription units, allowances, rates and cancellation terms remain unknown.

  • Public download and plan documentation does not establish signup completion, current charges or upgrade eligibility. No authenticated billing test was performed.
  1. Conductor is joining Meta, revised announcement

    Finding: The shutdown paragraph is struck through. Its replacement says development will continue; the FAQ says normal operation continues. Transfer to local and account export remain documented. The FAQ retains the no-sale/no-training statement but no longer sets deletion dates. The revision has no displayed date.

  2. Conductor 0.94.0 and 0.94.2

    Finding: The 0.94.0 release is listed November 9, 2026. Its page adds an undated 0.94.2 subsection recording removal of the Cloud shutdown banner. The parent entry also records model and bug-fix updates. November 9 is not an established date for the 0.94.2 subsection.

  3. Conductor pricing redirect

    Finding: The pricing URL redirects to the homepage, which advertises Cloud, desktop and mobile and links Get Conductor. No plan price table appears. This does not establish signup completion, upgrade eligibility or free-tier entitlements.

  4. Conductor Cloud FAQ

    Finding: Cloud access is described for Pro, Teams and Enterprise; Team signup is invite-only. The FAQ states no additional Cloud compute fee currently and plans usage-based compute pricing. Users bring subscriptions or keys. It also describes managed Vercel sandboxes and a planned self-hosted offering, without release dates. The mobile coming-soon wording is stale against the released iOS app.

  5. Configure model providers

    Finding: Model usage is billed through the selected provider account, subscription, usage pool or key. Conductor does not bill or resell it.