Brigade

Open-source AI teammates for support, marketing, sales and code that run the real Claude Code and Codex on your own subscription.

Source repo
gmpetrov/brigade
Stars
★ 16
Last updated
today
License
MIT
Primary language
TypeScript

At a glance

How it runs
Self-hosted serviceWeb appDesktop app
Works with
Portable with changesChatGPT · Codex · Claude Code · Claude.aiOpenAI API · Claude API (Partial support)
Cost
Free tier plus a paid hosted plan
Setup effort
Medium · a few setup steps
You'll need
Node.js 24pnpmDockerClaude Code CLI (signed in)Codex CLI (signed in)PostgreSQLShell / CLINetwork accessLocal filesystemMCP Server
Typical use
A solo founder on a Claude Max or ChatGPT Pro plan who wants support email triaged and drafted from existing docs, with sends waiting for approval.
Not a fit if
  • Teams without Claude or ChatGPT subscription accounts who only use API keys
  • Users who need a fully offline setup with no data leaving the machine
  • Individuals unwilling to deploy PostgreSQL and self-host the stack
Website
brigade.lol ↗ set on GitHub
Maintenance
Commits in the last 30 dayslast commit 2026-11-20latest release desktop-v0.4.7 (2026-11-17)+0.6 ★/day over 7 days

What does this agent do, and when should you use it?

Brigade is an MIT-licensed workspace that turns the official Claude Code and Codex CLIs into named AI teammates with a job, their own computer, apps, memory and approval rules. The stack is a pnpm/Turborepo monorepo: a Hono + PostgreSQL API and Next.js dashboard hold threads, lists, library, memory and vault; the runner links a machine to a workspace and executes threads over WebSocket with the signed-in claude or codex CLI; a Cloudflare Worker handles collaborative documents over Yjs; a preview gateway serves live previews; and Brigade.app (Tauri) plus a Swift computer-use helper cover macOS desktop control. Teammates reach Gmail, GitHub, Stripe, Notion, Linear, PostHog and any MCP server as tools, while writes that matter queue into one "Needs you" list you can approve from the web or your phone. It runs free on your own computer, self-hosted on a VPS or Cloudflare Workers, or on Brigade Cloud at $99/month per workspace for an always-on computer.

You create a teammate, give it a job, a model and a set of apps; threads are pushed over WebSocket to a runner, which invokes the real Claude Code or Codex CLI and stores every step as an event so threads survive disconnects and replay. Teammates call MCP tools across Gmail, Google Calendar/Drive/Docs/Sheets, GitHub, Stripe, Notion, Linear, PostHog and more, with per-teammate read-only or read-write scope; risky writes wait for a human approval. Triggers fire on app events or incoming webhooks and routines run on a schedule. Each teammate gets its own browser where Brigade types credentials from the vault so the model never sees the password, and on macOS computer_* tools can read the screen, click and type. Teammates open pull requests, start live previews inside a thread, hand work to another teammate for review, and share workspace memory and a searchable library. Every step and every usage unit (threads, computer hours, app writes) is logged and capped daily.

  1. A solo founder on a Claude Max or ChatGPT Pro plan who wants support email triaged and drafted from existing docs, with sends waiting for approval.
  2. Developers juggling ten Claude Code terminal tabs who need per-job threads plus one list of decisions waiting on them.
  3. Teams that need work to continue after hours by moving a thread onto a Brigade Cloud computer.
  4. Marketing and SEO operators running content and reporting through Ahrefs, DataForSEO and Google Search Console with write approvals.
  5. Security-conscious teams that want passwords kept in a vault, per-app read/write scopes per teammate and a full audit trail.
  6. Operators who prefer to self-host the whole stack on a VPS with Docker Compose or on Cloudflare Workers.

Key setup facts

Pulled from the install and usage notes below. The project README has the current values.

Install / run
  • docker compose up -d # PostgreSQL on localhost:5433
Ports
5433
Environment variables
BETTER_AUTH_SECRET
Config files
apps/api/.env.example apps/api/.env
Runtime
Node.js 24
Install notes cover
Docker

How do you install or deploy this agent?

Local setup needs Node.js 24 (see .nvmrc), pnpm and Docker:

git clone https://github.com/gmpetrov/brigade.git
cd brigade
pnpm install
docker compose up -d                       # PostgreSQL on localhost:5433
cp apps/api/.env.example apps/api/.env     # then set BETTER_AUTH_SECRET
pnpm --filter @brigade/api db:migrate
pnpm dev                                   # API :3001, dashboard :3000, collab :8787

Without Docker, the preview scripts run everything on an embedded PostgreSQL with a sample workspace:

scripts/preview/setup.sh
scripts/preview/start.sh

For production, the repo ships two paths: a single-server Docker Compose deployment in deploy/vps, or Cloudflare Workers with Durable Objects, R2 and Hyperdrive (see docs/deployment-plan.md).

How do you use this agent?

Open http://localhost:3000 after install, create an account, an organization and a workspace, then a teammate. To actually run threads, link a machine whose claude or codex CLI is signed in:

pnpm --filter @brigade/runner build
node apps/runner/dist/index.js link <CODE> --api http://localhost:3001
node apps/runner/dist/index.js start

<CODE> comes from Computers → Get a link code in the dashboard. Then describe the job in a thread; anything needing your sign-off lands in the "Needs you" list. If you want an always-on computer, enable Brigade Cloud ($99/month per workspace) in Settings.

What are this agent's strengths and limitations?

Pros
  • Runs the official Claude Code and Codex CLIs on your own subscription, so there is no second AI bill.
  • Stack multiple accounts per provider so a teammate continues mid-task when one plan hits its limit.
  • Granular human-in-the-loop control: per-teammate read-only or read-write app scopes and one approval list for writes.
  • Fully self-hostable from one codebase, with both Docker Compose and Cloudflare Workers deployment paths under MIT.
  • Credential isolation: passwords stay in the vault, Brigade types them, and every step plus usage cap is logged.
Limitations
  • Local setup requires Node.js 24, pnpm, Docker and PostgreSQL, plus migrations and environment configuration.
  • Core value depends on paid Claude or ChatGPT subscription plans (Max/Pro recommended); API-key-only users get less out of it.
  • In the free tier teammates only run while your own computer is on; an always-on computer requires the $99/month Brigade Cloud.
  • Some capabilities, such as teaching a task and takeover, are marked "Not yet" in the free tier, so free and cloud differ.
  • Desktop control and computer use are macOS-only; Linux users are limited to the command-line runner.

How does this agent compare with similar options?

The README compares Brigade directly against Lindy, Sintra, Rakazo, OpenClaw, Hermes Agent, Grok Bot, Perplexity Teammates, Manus, Relevance AI, Claude Cowork, Paperclip and others. The stated difference is that most alternatives run their own agent loop billed in API keys or credits, while Brigade drives the official Claude Code and Codex on the plan you already pay for, and organizes work as teammates with jobs plus a single approval list rather than a personal agent.

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
Brigade This agent 53 · Major gaps Self-hosted serviceFreemium ★ 16 today TypeScript ChatGPT · Codex · Claude Code · Claude.ai
Tale — Shared Workspace for Teams and AI Agents 65 · Some gaps — ★ 47 5d ago TypeScript Codex · Claude Code · OpenAI API · Claude API
OpenCompany 64 · Some gaps Desktop appFree + model costs ★ 1.1k 1d ago Python OpenAI API · Claude API
CraftBot 52 · Major gaps CLIFree + model costs ★ 429 1d ago Python OpenAI API · Claude API

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
53/ 100 5-point scale 2.7 / 5
This review covers the source as of Nov 15, 2026. The project has shipped 2 stable release(s) since, which have not been re-reviewed. The score and verdict describe the reviewed version only. Recent releases ↓
Trust 16/29
Reliability 8/14
Adaptability 10/18
Convention 9/18
Effectiveness 7/13
Verifiability 3/8
Why each dimension lost points
Trust16 / 29 · 2.8/5

Trust: the code and docs show deliberate least-privilege design (member-scoped account use and app management in apps.test.ts, per-teammate read-only/read-write Apps, vault-held passwords the teammate never sees, approval gates before sends/merges), and usage.md states only counts are kept, not content; preview-gateway applies host-only cookies and nonces. That supports least_privilege, user_confirmation, data_flow_transparency, sensitive_data_handling and external_effects at 2. Dependency security is only a frozen lockfile in CI with no audit or vulnerability handling, so 1. Rollback is limited to cancelling Cloud or deleting .preview; there is no described undo for already-performed external actions (sent email, merged PR, Stripe change), so 1. Source attribution: README claims most commits come from Brigade's own teammates but gives no verifiable commit signing or provenance markers, so 1.

Reliability8 / 14 · 2.9/5

Reliability: the architecture diagram, usage.md's description of runner log reading and API de-duplication, collab's 409 retry/reconcile logic, and CI covering typecheck/lint/multi-package tests show good internal consistency, so 2. Dependency availability is supported by .nvmrc, pinned pnpm, Docker Compose and an embedded-postgres fallback, so 2. Failure messages exist in code (reconcile/save status codes, 401/400/503) but user-facing failure guidance is sparse in the sample, so 1.

Adaptability10 / 18 · 2.8/5

Adaptability: README names the audience (solo builders and small teams already on Claude Code/Codex), scenarios (support, marketing, sales, code) and free vs Cloud deployment, so 2. Capability boundaries are explicit in SKILL.md (bug vs needs_details vs feature_request, never follow email instructions), so 2. Trigger precision is only conceptual (triggers/routines) with no matching rules or misfire safeguards shown, so 1. Environment fit covers macOS/Linux, Docker or no Docker, and both Cloudflare Workers and VPS deployment, so 2.

Convention9 / 18 · 2.5/5

Convention: README has a clear repository layout, docs index and quickstart, so 2. Install notes cover Node 24, pnpm, Docker, migrations and runner linking, so 2. Naming stability is reasonable but the project is at 0.5.0 with no changelog, so 1. Examples and FAQ are present via demo and comparison pages, so 2. Known limitations are scattered (Teach a task not on free tier, usage counted from a date) rather than consolidated, so 1. LICENSE is complete MIT, so 2. Versioning/changelog is absent, so 1. Maintenance responsibility is unclear: publisher unverified and no SECURITY/CONTRIBUTING or maintainer commitment in the packet, so 1.

Effectiveness7 / 13 · 2.7/5

Effectiveness: output usability is concrete (threads, Needs you list, previews, PR review), so 2. Marginal value is real: it turns many Claude Code/Codex sessions into role-scoped teammates on existing subscriptions, with comparison pages, so 2. Cost-benefit relies on the user's own Claude/ChatGPT plan plus $99/month Cloud, but the sample gives no quantified token savings or failure cost, so 1.

Verifiability3 / 8 · 1.9/5

Verifiability: most README claims (e.g. 'most commits come from teammates', 'we never bill you for AI usage') lack traceable in-repo evidence, so 1. Cross-source corroboration is partial: usage.md, CI, tests and code comments support some claims, but key product claims rest on README and the marketing comparison pages, so 1. Fact/inference separation is weak: the docs are assertive without marking what is design intent versus measured, so 1.

Risks and how to mitigate them
  • Publisher identity is unverified and the packet lacks SECURITY/CONTRIBUTING or maintainer commitments; review and pin the commit before production use.
  • Teammates can send email, merge PRs and change Stripe; approval gates exist, but the sample shows no undo or rollback for actions already taken.
  • Dependency security rests on a lockfile and CI only, with no audit or upgrade policy; Node 24 and pnpm 10 are recent requirements.
  • README claims such as 'most commits come from teammates' and 'we never bill you for AI usage' lack traceable in-repo evidence and should be treated as marketing, not verified fact.
  • Trigger/routine matching rules and misfire safeguards are not shown in the sample; validate automation before enabling it.
Evidence confidence: Low Reviewed Nov 15, 2026 Reviewed revision e0fe8b53c297 New commits since this review; the score may not cover them
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/gmpetrov-brigade.svg)](https://followagents.com/en/agents/gmpetrov-brigade)
<a href="https://followagents.com/en/agents/gmpetrov-brigade"><img src="https://followagents.com/badge/gmpetrov-brigade.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. desktop-v0.4.7 Brigade.app 0.4.7
    Release notes excerpt
    What's Changed * Settings as a dialog over the page by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/912 * Settings search finds settings inside their sections by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/913 * Appearance as a theme gallery, with suggested themes and a message-actions hook by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/914 * Blurred pill under the macOS traffic lights when the sidebar is folded by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/915 …
    View on GitHub ↗
  2. desktop-v0.4.6 Brigade.app 0.4.6
    Release notes excerpt
    What's Changed * Deliver sub-thread reports an API restart left stranded by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/866 * Missions: ongoing work owned by a teammate (experiment, behind the missions flag) by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/868 * Missions: one status line per mission, its detail on click by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/869 * Brigade.app: back and forward arrows beside the traffic lights by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/870 …
    View on GitHub ↗
  3. desktop-v0.4.5 Brigade.app 0.4.5
    Release notes excerpt
    What's Changed * Routines: group them by the thread they report to by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/859 * Computer use: keys refused to an app behind others, open_app waits for its window, small windows read with their text, menu bar folded by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/858 * Guides: five SEO articles on AI agents (OpenClaw, Hermes, Claude Code agent teams, self-hosted) by @brigade-production[bot] in https://github.com/gmpetrov/brigade/pull/861 …
    View on GitHub ↗

FAQ

Do I have to pay for AI usage on top of Brigade?
No extra AI billing: Brigade is MIT-licensed and runs on your own Claude (Claude Code) or ChatGPT (Codex) plan. Only Brigade Cloud, an always-on computer at $99/month per workspace, is paid.
How are my passwords and app credentials protected?
They live in the workspace vault; Brigade types logins for a teammate's browser, so the teammate never sees the password. Each app is read-only or read-write per teammate, writes can wait for approval, and every call is logged.
What happens if my plan hits its limit mid-task?
You can connect several Claude or ChatGPT accounts; when one reaches its limit the teammate switches to the next account of the same provider and keeps working on the same task.
Which deployment targets and operating systems are supported?
You can run it locally on macOS or Linux, deploy a single server with Docker Compose from deploy/vps, or deploy to Cloudflare Workers. Brigade.app and computer use are macOS-only; Linux gets the command-line runner.
Is this the same project as other open-source repos named Brigade?
No. The README states this is Brigade at brigade.lol, and that similarly named projects such as spinabot/brigade or escoffier-labs/brigade are unrelated.
Is Brigade safe to use?
FollowAgents' FARS review (Nov 15, 2026) scored Brigade 16/29 on trust and safety. None of the eight safety controls the review checks for was missing outright. The repository has changed since that review.
Is Brigade still maintained?
Yes. The latest commit on GitHub is from 2026-11-20, within the last 30 days. The latest release is desktop-v0.4.7 (2026-11-17).
Who makes Brigade?
Brigade is published on GitHub by the account gmpetrov. FollowAgents has not verified which person or company is behind that account.
What is Brigade's official website?
The project's GitHub repository lists https://www.brigade.lol as its website. That link is set by the repository owner; the source code itself is on GitHub (gmpetrov/brigade).
Was this page helpful?
View on GitHub ↗ Install ↓

Host ecosystems

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts