Skip to content

Read before entering site

Safe AI coding agents start with a site safety board

Foreman is a free, open-source desktop app for macOS and Windows that runs AI coding agents in parallel. Claude Code, Codex, Cursor and OpenCode each work in their own git worktree on their own branch, while you review diffs, answer questions and open pull requests from one window.

Every control on this page is real, says whether it's on by default or one you switch on, and says where it stops.

  • macOS & Windows
  • Apache-2.0
  • No account
  • No telemetry upload

Inspect the hazard

Pick a hazard. Watch it play out, then read the control that applies and the exact screen it puts in front of you.

Hazard 01 · inspection report

An agent pushes to origin

  1. What happensosaka has finished fix-login-redirect. Claude Code asks to test, commit and push.
  2. What Foreman doesApprovals are on, so the turn stops and Foreman spells the command out first.

Claude Code · osaka

Run 3 commands?

  • npm testRuns tests
  • git commit -am "Fix login redirect"Makes a commit
  • git push origin fix-login-redirectPushes to origin▲ Leaves this computer

NoAlways allowYes

  • You turn it onApprovals: Settings → Account → Claude Code tool approvals. Codex can stay in its sandbox, where git data is read-only.
  • DefaultNothing that leaves this computer has a keyboard shortcut, and Enter never answers a card.
  • DefaultThe agent pushes its own branch from its own worktree, never your checkout.

Where it stops: Out of the box Claude Code runs tools without asking. That is the default; turn approvals on if you want every push spelled out.

Hazard 02 · inspection report

An agent reads your API key

  1. What happenslisbon needs the Linear issue behind its task.
  2. What Foreman doesIt gets Linear tools through Foreman, at the level you chose. It never gets the key.

What lisbon's agent gets

  • Linear tools, Read only through Foreman
  • The Linear API key encrypted on this computer
  • The Slack bot token encrypted on this computer
  • Your E2B key, inside the sandbox read here, never sent in
  • GH_TOKEN from the GitHub connection, so it can push
  • DefaultLinear, Slack and Langfuse keys stay in Foreman's encrypted storage and are not passed to agents.
  • DefaultSandbox provider keys are read on your computer and never go into a sandbox.
  • DefaultSecrets are never command-line arguments, and repo MCP servers name variables, never values.

Where it stops: A GitHub connection gives agents GH_TOKEN on purpose, so they can push. And an agent runs as you: what your user can read, its shell can read.

Hazard 03 · inspection report

Two agents edit the same file

  1. What happensosaka (Claude Code) and lisbon (Codex) both open src/auth/login.ts.
  2. What Foreman doesEach edits its own copy, in its own worktree, on its own branch.

workspaces/osaka/src/auth/login.tsfix-login-redirect · +12 −3

workspaces/lisbon/src/auth/login.tsadd-rate-limit · +10 −1

You~/code/atlas/src/auth/login.tsmain · untouched

  • DefaultOne git worktree per workspace, each on its own branch. Your main checkout is never touched.
  • DefaultEach workspace gets its own block of ports, so two dev servers don't collide.
  • DefaultFlows and delegation never merge branches. Merging stays with you.

Where it stops: Two branches can still conflict when you merge them. Turn on the base-branch watch and Foreman tells you when one no longer merges; it never rebases for you.

Hazard 04 · inspection report

A phone you lost

  1. What happensYour phone was paired with Operate access, over your Tailscale network.
  2. What Foreman doesOn the computer, open Paired phones and press Revoke. It is signed out at once.

Paired phones Illustrative

  • iPhoneOperate · Private (Tailscale) · seen 2 h agoSigned out
  • iPadWatch · Same Wi-Fi · seen yesterdaySigned out
  • You turn it onPhone access is off until you turn it on. Pairing needs a one-time QR code and the same six-digit code on both screens.
  • DefaultEach device gets a level: Watch, Operate or Full control. The phone page offers nothing beyond Operate.
  • DefaultNo relay: same Wi-Fi connects directly, anywhere else only through your own Tailscale network.

Where it stops: To change a device's level you revoke it and pair again. The phone can't type into terminals at any level.

Hazard 05 · inspection report

A scheduled prompt gets changed

  1. What happensSomeone edits the text of the nightly-triage template.
  2. What Foreman doesThe schedule pauses before its next run, until you approve it again.

nightly-triage Paused.

Every day 09:00 · Europe/London · Claude Code

− Triage new issues and label them.
+ Triage new issues, label them and close duplicates.

Review… shows what changed and approves it again.

  • DefaultSaving a schedule approves its template, the repo's setup script, its checks and its MCP servers. Change any one and it pauses.
  • DefaultShared commands run only after you approve their text on this computer. Repo MCP servers need approval too.
  • DefaultUnattended runs are Claude Code only, and any approval they hit is denied and the turn stops.

Where it stops: Schedules run only while Foreman runs. A run more than 5 minutes late is skipped, never replayed.

Hazard 06 · inspection report

An agent runs rm -rf in your repo

  1. What happensThe agent in kyoto runs rm -rf src/legacy and takes 38 files with it.
  2. What Foreman doesRestore the turn: the files come back from Foreman's checkpoint.

rm -rf src/legacy38 files deleted

Restore files and conversation to before this message? Later changes are discarded.

38 files back

  • DefaultA checkpoint at the start and end of every turn: HEAD, index and the whole working tree, untracked files included.
  • You turn it onWith approvals on, a delete is marked Can't be undone before it runs.
  • You turn it onRead-only chats: Claude Code plan mode and the Codex sandbox. Not Cursor, OpenCode or ACP agents.

Where it stops: A checkpoint covers the workspace. Foreman runs tools as you, and plan mode is not an operating-system sandbox, so a command aimed outside the worktree is for approvals to catch.

Teardown

What does a Foreman approval actually show you?

The whole command, part by part, in plain words, with anything that leaves your computer marked and exactly what Always allow would add. Approvals are a setting: when they're on, this is the card.

You turn it on Settings → Account → Claude Code tool approvals

Claude Code · osaka · fix-login-redirect

Run 3 commands?

npm test && git commit -am "Fix login redirect" && git push --force-with-lease origin fix-login-redirect

Parts of the command

  1. npm testRuns testsOn this computer
  2. only if that succeeds
  3. git commit -am "Fix login redirect"Makes a commitOn this computer
  4. only if that succeeds
  5. git push --force-with-lease origin fix-login-redirectForce-pushes to origin▲ Leaves this computerCan't be undone

Always allow: commands starting with npm test, git commit or git push run without asking in this chat. That includes git push, which leaves this computer.

Tell the agent what to do instead (optional)

NoAlways allowYes

  1. A chained command is counted part by part, so nothing hides behind &&. The card docks above the message box and keeps what you were typing.

  2. The exact text the agent asked to run, before any summary.

  3. Runs tests. Makes a commit. Pushes to origin. One line per part, saying what it does.

  4. “Only if that succeeds” for &&; “then” and “given its output” for ; and |.

  5. A push, a pull request, an upload: anything that goes off your machine is marked before you answer.

  6. A force-push, a delete or a hard reset is flagged, so you read it twice. A plain push only leaves.

  7. The rule, where it applies (this chat, or the rest of an AI employee's run), what it still leaves out, and a warning when it covers something that leaves this computer.

  8. Tell the agent what to do instead, or just say no.

  9. Enter never answers the card, and nothing that leaves this computer has a keyboard shortcut. F6 moves to the card.

Illustrative card, built from the app's own wording. Hover or tap a number, or a line of the key.

Site safety plan

What keeps AI coding agents safe in Foreman?

Isolation first: every agent works in its own git worktree, and every turn can be rewound. Then the controls you choose: approvals, read-only chats and device levels. Nothing needs a Foreman account, and nothing is uploaded.

  • Every agent in its own worktree

    Its own checkout on its own branch. Agents can't trample yours.

    Default
  • Rewind any turn

    Checkpoints keep the whole working tree, untracked files included.

    Default
  • Read every diff

    Review each change before you commit, push or open a pull request.

    Default
  • Approvals you choose

    When on, a command is spelled out part by part, with what Always allow adds.

    You turn it on
  • Read-only when you want it

    Claude Code plan mode and the Codex sandbox. Those two only.

    You turn it on
  • Approve what a repo brings

    Shared commands, check commands and repo MCP servers run only after you approve them here.

    Default
  • Leaves this computer, marked

    Pushes, pull requests and posts are flagged on every approval and held action.

    Default
  • AI employees: outward actions

    Slack posts and Linear changes are held. Since 1.2.4 pushes and draft PRs go ahead; one line per employee holds them.

    Default
  • Outside agents over MCP

    Opt-in, capped by a level you pick. A level bounds what comes through Foreman, not the agent itself.

    You turn it on
  • Devices on a short leash

    Watch, Operate or Full control, a six-digit pairing code, instant revoke, no relay.

    You turn it on
  • No account, no telemetry upload

    There is no Foreman account, no cloud service and no upload.

    Default
  • Keys don't travel

    Integration keys are not passed to agents; provider keys never enter a sandbox.

    Default
A file diff with the list of changed files
Mandatory The changes list and a file diff. Every change is read here before it ships.

Exits

Does my code leave my computer?

Only through what you connect. Foreman has no account, no cloud service and no telemetry service, and it sends nothing else to any networked system unless you ask it to.

  1. Your agents talk to their own model providers, with your logins and keys.
  2. Services you connect: GitHub, GitLab, Linear, Slack, sandboxes and MCP servers.
  3. The update check: one unauthenticated request to api.github.com every few hours. GitHub sees your IP and version. You can turn it off.
  4. Phone and browser access, only once you turn it on and pair. Push alerts carry a generic title.

Diagnostics stay in your data folder. A bug report bundle is yours to save or attach, redacted.

Inspection tag · releases

Can I trust the download?

Releases are built by GitHub Actions from a version tag on main, never on a developer's computer.

macOS
Signed and notarized
Every file
SHA-256 in SHA256SUMS.txt
Source
All of it, Apache-2.0
Releases
31 since 27 October
Check a download
sha256sum -c SHA256SUMS.txt --ignore-missing

Rest of the site

What else does Foreman do, and how does each part stay safe?

Foreman also runs automations, flows, remote sandboxes and AI employees, and lets you steer from your phone. Each one comes with its own guard rail.

  • Automations and schedules

    A schedule pauses when what it runs changes, and two failed runs in a row pause it too.

    Claude Code only, while Foreman runs

  • Flows that loop until green

    Loops stop on facts, like checks passing, never on an agent saying it's done. Gates wait for you. 11 ship built in.

    Experimental · off by default

  • Remote sandboxes: E2B, Daytona, Vercel, Docker

    Your own account, your own limits. Provider keys stay on your computer. Needs a GitHub remote; not Cursor.

    Experimental · off by default

  • AI employees

    A changed definition pauses the employee until you approve it again, with each difference shown.

    15 job templates

Remote control

Run it from anywhere, with three levels

Pair your phone with a QR code to approve, steer and comment on diffs, live. Away from home it connects over your own Tailscale network, with no Foreman relay. A browser on another computer can open the whole app.

  • WatchSee workspaces, chats, diffs and checks. Change nothing.
  • OperateAlso send messages, answer approvals, run checks and archive.
  • Full controlAlso settings, repositories, keys and integrations. For a browser.

Today: pair your phone with a QR code and add the phone view to your Home Screen.

Other sites

How is Foreman different from Conductor and T3 Code?

Foreman is open source under Apache-2.0, so every control on this page can be read in the code, and it runs on Windows as well as macOS. Conductor is a closed-source Mac app; next to T3 Code, Foreman adds flows, AI employees, schedules and a no-telemetry promise.

A dash means not listed on their site

Foreman compared with Conductor and T3 Code
FeatureForemanConductorT3 Code
Open sourceApache-2.0NoMIT
macOSYesYesYes
WindowsYesMac onlyYes
One git worktree per agentYesYesYes
AgentsClaude Code, Codex, Cursor, OpenCode, any ACP agentClaude Code, Codex, CursorClaude Code, Codex and more
Flows that loop until checks pass, with human gatesYes——
AI employees with git-backed memoryYes——
Scheduled automations and auto-fix CIYes——
Your own cloud sandboxes (E2B, Daytona, Vercel, Docker)Experimental——
Scriptable CLI, --json on every commandYes——
Approve from your phoneYes—Yes
No account, no telemetry uploadYes——
From each project's own site as of November 2026. “—” means it isn't listed there, not that it's missing. Spot a mistake? Tell us.

Site induction

Get started in a minute

Download the app, add a git repository, and give each task its own workspace. Switch on approvals if you want every command spelled out.

  1. Download Foreman

    Free for macOS (Apple Silicon, signed and notarized) and Windows. You need Git on your PATH.

    Download Foreman
  2. Add a repo, sign in an agent

    Claude Code, Codex, Cursor or OpenCode, in Settings → Agents, with your own login or key.

  3. Create workspaces

    One per task, each a fresh worktree on its own branch. Read the diff, then open the pull request.

From a terminal
foreman run fix-a-bug "the login page loops after sign-in" --repo atlas --wait
foreman approvals

# Start a workspace from a template and wait for the turn, then see what is waiting for you. CLI reference

Toolbox talk

Questions developers ask

Found a hazard we missed? Open an issue or read the guide.

What is Foreman?

Foreman is a free, open-source desktop app for macOS and Windows that runs AI coding agents in parallel. Claude Code, Codex, Cursor and OpenCode each work in their own git worktree on their own branch, while you review diffs, answer questions and open pull requests from one window.

Is Foreman free and open source?

Yes. Foreman is free and licensed under Apache-2.0, with the full source on GitHub at github.com/elite-build/foreman. There is no paid tier, no account and no Foreman cloud service.

Which coding agents does Foreman support?

Foreman runs Claude Code, Codex, Cursor and OpenCode, plus any agent that speaks the Agent Client Protocol (ACP). Each agent signs in with your own subscription or API key, set in Settings → Agents.

How do I run multiple Claude Code agents in parallel?

Add a git repository to Foreman and create one workspace per task. Each workspace is a separate git worktree on its own branch, so several Claude Code sessions edit the same repository at once without conflicts. Tiles shows them side by side.

Why does each agent need its own git worktree?

A git worktree is a second checkout of the same repository on a different branch. Giving each agent its own worktree means agents never overwrite each other's files, every change lands on a reviewable branch, and your main checkout stays untouched.

Does my code leave my computer?

Only through the agents you choose. Foreman runs locally with no account and no telemetry upload; your agents talk to their own model providers. In your own workspaces, an approval marks any command that leaves your computer before it runs, and AI employees hold Slack posts and Linear changes for you.

Do coding agents need my approval for every command?

Only if you want them to. Approvals are a setting: with Claude Code tool approvals on (Settings → Account), each command is spelled out part by part before it runs, anything that leaves your computer is marked, and Always allow shows exactly what it adds. Every agent still works in its own worktree, and checkpoints rewind any turn.

Which operating systems does Foreman run on?

Foreman runs on macOS (Apple Silicon, signed and notarized) and Windows. You need Git on your PATH and a login or API key for at least one agent.

Can Foreman run coding agents on a schedule?

Yes. Any saved prompt template can run hourly, daily, weekly or monthly in your time zone, and opt-in auto-fix sends failing CI logs to your agent. Scheduled runs use Claude Code and happen while Foreman is running, which can be without a window or from login.

Can I control Foreman from my phone?

Yes. Pair a phone with a QR code to answer approvals, steer agents, comment on diffs and read terminals, live. On the same Wi-Fi it connects directly, and anywhere else through your own Tailscale network. Native iOS and Android apps are coming soon.

Can coding agents run in a cloud sandbox like E2B?

Yes, as an experiment. A workspace's worktree and agent can run in a sandbox on your own E2B, Daytona, Vercel Sandbox or Docker account, for Claude Code, Codex and OpenCode. Provider keys stay on your computer and never go into the sandbox.

Which integrations does Foreman have?

GitHub (pull requests, merges, CI and reviews), GitLab, Linear, Slack and Langfuse, plus MCP. Each connection has a default per service, each repository can pick its own, and you choose what agents may do with it: nothing, read, or read and write.

How is Foreman different from Conductor?

Foreman is open source under Apache-2.0 and runs on Windows as well as macOS, while Conductor is a closed-source Mac app. Foreman also drives OpenCode and any ACP agent, and adds flows that loop until checks pass, AI employees and a scriptable command line.

How is Foreman different from T3 Code?

Both are open-source GUIs for coding agents. Foreman adds flows (agent graphs that loop until tests pass, with human gates), AI employees with a git-backed memory, a foreman CLI where every command takes --json, phone approvals and remote sandboxes.

Can I script Foreman or use it in CI?

Yes. The foreman command starts workspaces, sends messages, answers approvals, opens and merges pull requests and runs flows. Every command takes --json and exit codes are the same for every command.

Safety committee · open to all

Inspect the code yourself.

Every control on this page is in a public repository, 31 releases and three weeks old. Read it, star it so you hear about the next release, fork it, or fix something on a good first issue.