Skip to content

Deal me in.

The best AI coding agent setup is a whole crew. Deal yours.

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.

  • macOS & Windows
  • Apache-2.0
  • No account
  • v1.2.4

The job

Loadout for fix flaky tests

  1. DeveloperClaude CodeAgent · Anthropic

    Writes the fix. Flow steps that change code run on Claude Code, the one agent Foreman runs unattended.

  2. ReviewerCodexAgent · OpenAI

    Reads the diff read-only, inside its own sandbox. A flow’s review steps can run on Codex.

  3. Review personaqa-engineerFlow persona · reviewer

    What the reviewer wears: it finds what breaks, never fixes it, and ends with findings.

  4. At the gateYouThe foreman · one of one

    Quarantine the test, keep the fix, or send it back with a note. Nothing ships before you look.

Plays as the built-in flow flaky-test-triage. Flows are experimental and off by default: Settings → Experimental.

Run them side by side in Foreman, each in its own worktree.

Agents · the crew board

Where does an agent card go to work?

To its own lane. Foreman gives every agent a git worktree on its own branch, so Claude Code, Codex, Cursor and OpenCode work one repository at the same time without stepping on each other's files.

  • Built and load-tested for 10 to 12 agents at once
  • Every lane ends as a diff you read before it ships
  • Checkpoints rewind any turn, untracked files included
Four Foreman workspaces side by side in Tiles: a failed turn, a finished fix and two questions from AI employees waiting for an answer
Tiles The real app: several workspaces side by side, each in its own worktree.

Tap a hat or a work order. Each agent gets its own lane: a git worktree on its own branch.

Work orders
  1. 01osaka

    Open lane · waiting for a hat

    fix-login-redirectClaude Code

    +12 −3 · ready for review

  2. 02lisbon

    Open lane · waiting for a hat

    add-rate-limitCodex

    +10 −1 · ready for review

  3. 03kyoto

    Open lane · waiting for a hat

    write-testsCursor

    +48 −0 · ready for review

  4. 04lima

    Open lane · waiting for a hat

    update-docsOpenCode

    +6 −2 · ready for review

4 worktrees open · 4 signed off

Away from the desk? You still hold the gate

Agents sharing a folder: none. Each agent works in its own git worktree.

Personas · the drafting desk

What does a persona card play as?

A step in a flow. A flow is a graph of agent steps with loops that stop on facts, like the checks passing or a review coming back clean, and gates where you decide. Personas say who does each step and what it may write.

  • 11 built-in flows, 10 personas
  • What a persona may write is enforced by Foreman, not the prompt
  • Experimental and off by default: Settings → Experimental
Start from a built-in flow

⁠, ⁠, ⁠, ⁠, ⁠.

Tap any underlined part to change it.

Matches the built-in flow fix-until-green.

Ready · 4 rooms · 2 loops · 1 gate

IN · {{task}}OUT · done01implementdeveloper · edits the worktreeTURN · CLAUDE CODE02checksnpm test · tsc · lintCHECKS · NO AGENT03reviewqa-engineer · read-onlyREVIEW · CLAUDE OR CODEX04ship-checkyou · ready to ship?GATE · YOU≤ 4 roundson failed · 45 min≤ 2 roundson findingssent backJOB
IN · {{task}}OUT · done01implementdeveloperTURN · CLAUDE CODE02checksnpm test · lintCHECKS · NO AGENT03reviewqa-engineerREVIEW · CLAUDE OR CODEX04ship-checkyou decideGATE · YOU≤ 4 roundson failed≤ 2 roundson findingssent backJOB
.foreman/flows/fix-until-green.tomlSimplified for this page
# .foreman/flows/fix-until-green.toml · simplified for this page
name = "Fix until green"
description = "Skip the plan, the developer builds, the checks run until green, at most 4 rounds, the QA engineer reviews until clean, and I approve before it ships."

[inputs]
task = "What to build or fix"

[start]
step = "implement"

[steps.implement]
kind = "turn"                  # code steps run on Claude Code
persona = "developer"
loadout = 1
prompt = "{{task}}"

[steps.checks]
kind = "checks"                # the repo's approved check commands, never an agent

[steps.fix]
kind = "turn"
same_tab = "implement"
prompt = "Address what failed or what you were sent back for, and run the checks again."
context = ["checks.results"]

[steps.review]
kind = "review"                # read-only: Claude Code or Codex
persona = "qa-engineer"
                               # its default loadout 3 is Codex
context = ["diff"]

[steps.address]
kind = "turn"
same_tab = "implement"
context = ["review.findings"]

[steps.ship-check]
kind = "gate"                  # you: Continue, Send back or Stop
show = ["diff", "checks.results", "review.findings"]
ask = "Is this ready to ship?"

[edges]
implement  = "checks"
checks     = { passed = "review", failed = "fix" }
fix        = "checks"
review     = { clean = "ship-check", findings = "address" }
address    = "checks"
ship-check = { continued = "done", sent-back = "address", stopped = "done" }

[loops.fix-until-green]
over = ["checks", "fix"]
at_most = { iterations = 4, minutes = 45 }
iteration = "same-tab"         # or "compact-when-full" or "fresh-with-summary"

[loops.review-until-clean]
over = ["review", "address", "checks"]
at_most = { iterations = 2 }

[after]
workspace = "keep"             # keep | archive | delete-unchanged
Project
atlas
Flow
fix-until-green
Drawn by
the assistant
Checked by
you
Revisions
  1. A first drawn

Scale: 1 step = 1 room · persona hats drawn for this page

Site log fix-login-redirect

  1. 00:08developer: moved the redirect after setSession() · +12 −3
  2. 00:45checks: ✕ auth/redirect.test.ts › keeps returnTo after sign-in
  3. 01:22developer (fix, same tab): kept returnTo on the refresh path · +4 −1
  4. 01:59checks: ✓ 214 tests · types · lint
  5. 02:36qa-engineer: clean, no findings
  6. 03:13ship-check: waiting for you · ready to ship?

Illustrative run of the sentence above. In Foreman the Flow tab draws it live and the gate waits in Needs you. Flows are experimental: turn them on in Settings → Experimental → Flows.

Legend

  • A room is a step: a turn, checks, a review, a gate or a PR
  • A loop, with its bound and the fact that sends it round
  • Sent back from your gate, with your note
  • Your gate is a door: nothing passes until you open it
  • The job, walking from room to room on a run

Employees · the roster

Can you keep an AI employee on the team?

Yes. Hire from 15 job templates. Each employee gets a role in plain words, a hard hat and access you approve, works in a fresh workspace each run, and remembers through the organization's own git memory repository.

Open roles 15

Engineering

Design

Product

Data

Writing

Marketing

Customers

Hat rule A hat says who, never state. If a job's colour is taken, the new hire wears the least-worn colour instead.

Slack posts and Linear changes start held for you. Since v1.2.4 an employee pushes and opens draft pull requests on its own; this switch is the one-line rule that holds those too.

Your team 3

  • You

    The person · one of the team

    Approve each plan and each hire, and anything you've asked the team to hold.

  • Priya Lead

    Program manager

    Plans every project, proposes who it needs, reviews each part and accepts the result.

    Starts
    When you @Priya
    Access
    Writes project filesNo shellSlack · held
  • Quinn

    Staff QA engineer

    Reviews what others build and says what breaks.

    Starts
    When you askWhen review is requested
    Access
    Write · atlasNo shellPush & PR · heldSlack & Linear · held

Since 1.2.4 employees push and open draft pull requests without waiting; Slack posts and Linear changes are still held for you. Unattended runs use Claude Code.

You · the gate

Can you play your hand from your phone?

Yes. Pair a phone with a QR code to answer approvals, plans and flow gates, steer agents and comment on diffs, live: directly on the same Wi‑Fi, and anywhere else over your own Tailscale network. There is no Foreman relay.

  • WatchSees everything, changes nothing.
  • OperateAnswers approvals, plans, questions and gates; steers and comments on diffs. Terminals are read-only.
  • Full controlAlso settings, repositories and keys. Meant for a browser on another computer.

Native iPhone and Android apps are on the way.

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

Illustrative A flow waits at your gate. Nothing ships before you look.

The table

What else is on the table?

Four more cards: the tools your crew carries, sandboxes it can work off-site in, a night shift, and the real app as built. Play one to lay it down.

Play: Tool belt Card 1 of 4

Which tools can the crew carry?

GitHub, GitLab, Linear, Slack, Langfuse and MCP. Each connection gets the agent access you choose: nothing, read, or read and write.

  • Slack is outbound only
  • Langfuse is read-only
  • Auto-fix answers failing CI, within a budget
  1. GitHub
  2. Linear
  3. Slack
  4. Langfuse
  5. MCP
  6. Sandboxes
  7. GitLab

Connections belong to the repository, so every new workspace in atlas starts with this belt.

7 of 7 tools on the belt

The tags on the belt

Each tool's tag says what the agent can do with it and what stays held for you. Change a setting and the run below changes too.

  • Wrench · gh or a token

    GitHub

    Auto-fix CI

    Commit, push and open draft PRs. Merge, or auto-merge when checks pass. Watch CI and reviews. Start from an issue or PR link. Open PRs in GitHub or Graphite.

    A failing CI run waits on Home with Fix CI in… for you to press.

    Auto-fix sends the failing log to the tab you picked: 2 fixes a commit, 5 a PR. Never for review comments.

    Agents in repositories that use it get GH_TOKEN. Turn approvals on to see each push first.

  • Level · an API key

    Linear

    What agents may do

    Agents get no Linear tools. Foreman still reads issues for New workspace.

    Agents can search and read issues, projects and comments.

    Agents can also create and update issues and comment.

    Nothing goes to Linear from an agent.

    They can't create, edit or comment.

    Issue text is written by others, so treat this like any tool that changes things.

    Foreman doesn't change issue status itself.

  • Megaphone · a bot token

    Slack

    What agents may do
    Rule: PR ready → message the reviewer

    Foreman sends your notification rules. Agents can't post.

    Agents may also post to #launches, the channel you allowed.

    Messages to anyone but you wait for your Send.

    This rule has your consent; change any part of it and it asks again.

    Outbound only. Never pings @here or @channel.

  • Magnifier · a key pair

    Langfuse

    What agents may do

    Agents get no Langfuse tools.

    Agents get search_observations, get_trace and get_observation.

    They can't change anything in Langfuse.

    Langfuse Cloud EU, US or HIPAA, or your own host over HTTPS.

  • Power strip · built in

    MCP

    Let agents start workspaces

    Every workspace agent gets foreman-app: its status, its checks, a preview, its own chat history.

    No agent can start a workspace or queue a message.

    Agents can hand tasks to other workspaces in atlas and read the replies. Other repositories are refused.

    MCP servers a repository proposes start only after you approve them.

  • Toolbox · experimental

    Sandboxes

    Where the workspace runs

    E2B · Hosted · 1-hour sessions on Hobby · 2, 4 or 8 vCPUs.

    Daytona · Hosted · 24-hour sessions · 1, 2 or 4 vCPUs.

    Vercel Sandbox · Hosted · 45-minute sessions on Hobby · 2, 4 or 8 vCPUs.

    Docker · This computer, or another over SSH · No session limit · 1 to 8 vCPUs.

    Off by default. Provider keys never enter the sandbox. Needs a GitHub remote. Claude Code, Codex and OpenCode, not Cursor.

    Turns keep going after you quit, up to the provider's session limit.

  • Pliers · an access token

    GitLab

    gitlab.com or self-managed

    Merge requests and issue links work like GitHub's, on your own GitLab host.

    Fewer events: employee triggers don't see failed pipelines, approvals or requested changes.

    foreman gitlab add https://gitlab.example.com

Example run: ENG-123, start to finish

8 of 8 steps work with this belt. Missing tools grey out their steps.

  1. Linear

    Starts from ENG-123, “Login loops after sign-in”. The agent reads the issue and its comments.foreman run issue ENG-123

  2. Workspace

    Workspace osaka runs in a E2B sandbox, on eng-123-login-loop. Provider keys stay on your computer.

  3. Langfuse

    Reads the trace of the failing POST /login: the redirect fires before the session is saved.

  4. MCP

    Checks itself with foreman-app. Delegation is off, so it writes the regression test too.

  5. Pull request

    Pushes the branch and opens draft PR #42 on GitHub.foreman pr create atlas --draft

  6. CI

    CI fails: Tests, 1 failing. Foreman saw it, so you don't have to refresh.

  7. Back to green

    Auto-fix is off: Home shows CI failed and you press Fix CI in…. Checks pass.

  8. Slack

    Drafts “PR #42 is ready for review” to @maya, the reviewer. It waits for your Send.

Merging stays with you: press Merge, or foreman pr merge atlas --auto to merge when the checks pass. Linear’s own GitHub integration moves ENG-123 along. Illustrative run; the names are made up.

Play: Off-site Card 2 of 4

Can a card keep playing after you close the laptop?

Yes, as an experiment. A workspace can run in a sandbox on your own E2B, Daytona, Vercel Sandbox or Docker account, and its turn goes on after you quit, until it finishes or meets the provider's session limit.

  • Provider keys never enter the sandbox
  • Claude Code, Codex, OpenCode; not Cursor
  • Needs a GitHub remote. Off by default

New workspace · atlas

Workspace

Runs on

Sandbox
E2B
Size
vCPU
Starts from
A fresh clone, set up when it’s made
Sleeps after
20 min idle
Agent sign-in
Claude setup token

Your Claude subscription sign-in stays on this computer. In a sandbox Claude uses a setup token from claude setup-token, or an API key.

On E2B a session ends after 60 minutes. Foreman continues the turn when you’re around; with Foreman closed, the agent stops at the limit and the workspace sleeps.

Remote workspaces run on E2B. Experimental.

This computer

Foreman · atlasZzz

Vault

  • Provider keystays here
  • GitHub tokenstays here
  • Agent sign-instays here
Property line · provider keys stop here
  1. Clone from GitHub The repository comes from its GitHub remote. GitHub’s token goes in as that one command’s environment.
  2. Install the agent host Foreman puts its small agent host in the sandbox, so the agent can run there without your computer.
  3. Set up The install script and the repository’s setup steps run, as they would in a local worktree.
  4. Run The agent takes the turn in the sandbox. Your computer only coordinates.

Session · not started

Turn · waiting to ship

  1. Pick a workspace and a container, then Create. Closing the laptop quits Foreman: remote turns go on in their sandboxes until they finish or reach the session limit, and their chats catch up when you open Foreman again.

This month · no sandbox time yet

Play: Night shift Card 3 of 4

Can the crew play while you sleep?

Yes, on Claude Code. Save a prompt as a template and Foreman runs it hourly, daily, weekly or monthly in your time zone, while Foreman runs: with a window, without one, or from login.

  • A sleeping computer isn't woken
  • A run 5 minutes late is skipped
  • Seven templates ship

The seven templates that ship 7

Saved prompts yours · blanks written out

Blanks fill in order: foreman run fix-a-bug "the login page loops". A scheduled run can't stop to ask, so only templates without {{blanks}} go on a schedule. Repository templates live in.foreman/templates and travel with the code.

Schedule slipNo. 0042

nightly-triage on atlas

Every
in Europe/London
Runs on
Claude Code, a Sonnet by default
Max
30 minutes5 to 480
After
Approved
its text, the repo's setup, checks and MCP servers

Runs every day at 00:30 (Europe/London) · stops after 30 minutes · keeps the workspace

foreman schedules add nightly-triage --repo atlas --every daily --at 00:30 --tz Europe/London

Play: As built Card 4 of 4

What does a flow look like in the real app?

Like this: the real flow studio, with a built-in flow's steps, loops and gate drawn as a graph and the assistant under it. Click the stamp; hover the crew.

research-then-implement · 7 steps · 2 loops · 1 gate
A flow's graph of steps, loops and a gate, with the assistant under itThe drawing's stamp. Press to stamp it again.
  • I plan the change. I never touch the files.
  • I build it, then fix it until the checks pass.
  • I find problems. Fixing them isn't my job.
  • Nothing ships until I open the gate.
Fig. 1The real flow studio: a built-in flow's steps, loops and gate, with the assistant under it.

Other tables

How is Foreman different from Conductor and T3 Code?

Foreman is open source under Apache-2.0 and runs on Windows as well as macOS, while Conductor is a closed-source Mac app. Next to T3 Code, Foreman adds flows that loop until checks pass, AI employees with git-backed memory and a CLI where every command takes --json.

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.

Cut, deal, play

Get started in a minute

Download the app, point it at a git repository, and give each task its own workspace.

  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

    Any git repository. Sign in 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. Deal the hand, then read every diff before it ships.

Prefer a terminal? Same deal, one line.
foreman run fix-a-bug "the login page loops after sign-in" --repo atlas --wait

# Start a workspace from a template and wait for the turn. CLI reference

House rules

Questions developers ask

Something missing? 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.

Pull up a chair. The game is open source.

Foreman is three weeks old: 31 releases since 1.0.0 on 27 October, built in the open under Apache-2.0. Star it to hear about the next one, fork it, or pick up a good first issue and take a seat at the table.