mergecrew

KO | EN

gitlite — search

#large-language-models #machine-learning #nodejs #video #pdf #chatbot #agents #security #html #react-native

(★ 9)

Open-source agentic SDLC platform: an autonomous product team that ships PRs daily, with a human approval gate before production.

파일 탐색기

ZIP 다운로드

- - bug_report.md
  - config.yml
  - feature_request.md
- - ci.yml
  - compose-smoke.yml
  - deploy-services.yml
  - deploy-vm.yml
  - deploy-web.yml
  - e2e-loop.yml
  - publish-runner-images.yml
- dependabot.yml
- FUNDING.yml
- PULL_REQUEST_TEMPLATE.md
- - nest-cli.json
  - package.json
  - tsc.err
  - tsconfig.json
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
  - tsconfig.test.json
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
  - vitest.config.ts
- - package.json
  - tsconfig.json
- - package.json
  - README.md
  - tsconfig.json
- - package.json
  - tsconfig.json
  - tsconfig.test.json
- - package.json
  - tsconfig.json
  - tsconfig.test.json
- - next-env.d.ts
  - next.config.mjs
  - package.json
  - postcss.config.mjs
  - tailwind.config.ts
  - tsconfig.json
  - vercel.json
- - package.json
  - tsconfig.json
- - 01-vision.md
  - 02-scope.md
  - 03-personas.md
  - 04-user-journeys.md
  - 05-features.md
- - 01-principles.md
  - 02-information-architecture.md
  - 03-key-screens.md
  - 04-realtime-timeline.md
  - 05-bureau-design-system.md
- - 01-overview.md
  - 02-domain-model.md
  - 03-multi-tenancy.md
  - 04-agentic-runtime.md
  - 05-ai-abstraction-layer.md
  - 06-workflow-engine.md
  - 07-vcs-adapter.md
  - 08-deploy-adapter.md
  - 09-data-model.md
  - 10-api-surface.md
  - 11-security.md
  - 12-skill-sdk.md
  - 13-runner-isolation.md
  - 14-runner-microvm-decision.md
- - 01-overview.md
  - 03-credit-and-rate-handling.md
  - 04-observability.md
  - 05-operator-runbook.md
  - 06-deploy-targets-cookbook.md
  - 07-telemetry.md
  - 08-monthly-spend-cap.md
  - 09-dry-run.md
  - 10-blast-radius.md
  - 11-risk-score-gate.md
  - 12-rollback.md
  - 13-spend-forecast.md
  - 14-anomaly-digest.md
  - 15-evals.md
  - 16-self-host-runbook.md
  - 17-install-telemetry.md
  - 18-multi-agent.md
  - 19-vm-deploy.md
  - 20-github-app-setup.md
  - 22-runner-images.md
  - 23-runner-network-policy.md
  - 24-stack-node.md
  - 25-stack-python.md
  - 26-stack-java.md
  - 27-stack-go.md
  - 28-stack-ruby.md
  - 29-stack-dotnet.md
  - 30-runner-kubernetes.md
  - 31-runner-fargate.md
  - 33-pause-resume.md
  - 34-runner-agent.md
  - 35-runner-fargate-byo.md
  - 36-runner-github-actions.md
  - 37-runner-docker-sock-security.md
- - 0001-adopt-adr-format.md
  - 0002-per-org-runner-profile.md
  - 0003-runner-agent-long-poll.md
  - 0004-runner-agent-token-storage.md
  - 0005-per-profile-queues.md
  - 0006-trusted-org-gating.md
  - 0007-byo-cloud-credentials.md
  - 0008-default-profile-none.md
  - 0009-byo-agent-as-remote-sandbox-driver.md
  - README.md
- 00-quickstart.md
- 04-roadmap.md
- openapi.json
- README.md
- - alb.tf
  - cluster.tf
  - ecr.tf
  - iam.tf
  - locals.tf
  - oidc.tf
  - outputs.tf
  - README.md
  - security-groups.tf
  - services.tf
  - variables.tf
  - versions.tf
- - Dockerfile.api
  - Dockerfile.orchestrator
  - Dockerfile.runner
  - Dockerfile.runner-agent
  - Dockerfile.runner-dns
  - Dockerfile.runner-egress-proxy
  - Dockerfile.web
  - Dockerfile.worker-cron
  - runner-entrypoint.sh
- - deploy-dev.yml
  - README.md
- - README.md
- - README.md
  - server.mjs
- - README.md
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
  - tsconfig.test.json
- - package.json
  - prisma.config.ts
  - tsconfig.json
  - vitest.config.ts
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
  - vitest.config.ts
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
  - tsconfig.test.json
- - package.json
  - tsconfig.json
  - tsconfig.test.json
- - package.json
  - README.md
  - tsconfig.json
- - package.json
  - tsconfig.json
  - tsconfig.test.json
- - package.json
  - tsconfig.json
- - package.json
  - tsconfig.json
- - budget-exhaust.ts
  - force-rate-limit.ts
  - redis-restart.ts
  - sse-disconnect.ts
- - capture.ts
  - README.md
  - routes.ts
- bootstrap.ts
- check-no-raw-sql.mjs
- deploy-pull.sh
- deploy.sh
- dogfood-bake-report.ts
- e2e-local.sh
- local-reset.sh
- provision-egress-network.sh
- quickstart-demo-run-smoke.sh
- quickstart-smoke.sh
- sse-replay-test.ts
- telemetry-preview.ts
- - pyproject.toml
  - README.md
.dockerignore
.env.example
.gitignore
.npmrc
.nvmrc
.prettierrc
CODE_OF_CONDUCT.md
CONTRIBUTING.md
docker-compose.full.yml
docker-compose.prod.yml
docker-compose.yml
eslint.config.mjs
LICENSE
package.json
pnpm-lock.yaml
pnpm-workspace.yaml
README.md
SECURITY.md
SPONSORS.md
tsconfig.base.json
turbo.json

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

> Autonomous product team in a box: every day, `mergecrew` specifies, designs, builds, deploys to dev, scans for bugs, and hands you a digest to approve before anything reaches production.

[![CI](https://github.com/mergecrew/mergecrew/actions/workflows/ci.yml/badge.svg?branch=main)](https://github.com/mergecrew/mergecrew/actions/workflows/ci.yml)
[![GitHub stars](https://img.shields.io/github/stars/mergecrew/mergecrew?style=flat&logo=github&color=yellow)](https://github.com/mergecrew/mergecrew/stargazers)
[![license: Apache 2.0](https://img.shields.io/badge/license-Apache%202.0-blue.svg)](https://raw.githubusercontent.com/mergecrew/mergecrew/main/LICENSE)
[![status: alpha](https://img.shields.io/badge/status-alpha-orange.svg)](https://raw.githubusercontent.com/mergecrew/mergecrew/main/)
[![docs](https://img.shields.io/badge/docs-spec-green.svg)](https://raw.githubusercontent.com/mergecrew/mergecrew/main/docs/README.md)
[![Discussions](https://img.shields.io/badge/discussions-ask-blueviolet?logo=github)](https://github.com/mergecrew/mergecrew/discussions)
[![PRs welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](https://raw.githubusercontent.com/mergecrew/mergecrew/main/CONTRIBUTING.md)

Mergecrew is the open-source platform for running an **agentic software development lifecycle** — multi-agent product teams that operate on a daily cadence against your real repo, deploy automatically to your dev environment, and require an explicit human approval before anything ships to prod.

```mermaid
flowchart LR
 Cron([Cron trigger no ticket needed]) --> Plan[Planner spec the change]
 Plan --> Code[Coder build the diff]
 Code --> Draft[Open draft PR]
 Draft --> Rev{LLM reviewer}
 Rev -- request_changes --> Code
 Rev -- approve --> Ready[Post review + mark ready-for-review]
 Ready --> Dev[Deploy to dev]
 Dev --> Scan[Bug scan preview env]
 Scan --> Digest[Daily digest]
 Digest --> Human{Human approval}
 Human -- approve --> Prod[Production]
 Human -- reject --> Stop([Stop])

classDef trigger fill:#fef3c7,stroke:#d97706,color:#92400e
    classDef agent fill:#dbeafe,stroke:#2563eb,color:#1e3a8a
    classDef gate fill:#fee2e2,stroke:#dc2626,color:#991b1b
    classDef ship fill:#dcfce7,stroke:#16a34a,color:#14532d
    classDef vcs fill:#ede9fe,stroke:#7c3aed,color:#4c1d95
    class Cron trigger
    class Plan,Code,Rev,Scan agent
    class Human gate
    class Dev,Digest,Prod ship
    class Draft,Ready vcs
```

> **Try it in 5 minutes:** `docker compose -f docker-compose.full.yml up`, open `http://localhost:3000`. You land directly in the seeded `demo-saas` project — a guided coachmark tour walks you through the completed sample run, then drops you in the wizard when you click **Set up your own project →**. Full walkthrough: [docs/00-quickstart.md](docs/00-quickstart.md).

### What it looks like

<table>
 <tr>
 <td width="50%">
 <a href="docs/assets/screenshots/02-today.light.desktop.png">
 <img src="docs/assets/screenshots/02-today.light.desktop.png" alt="Today dashboard — post-login hub with welcome card, project list, and recent activity feed." />
 </a>
 Today — post-login dashboard. Welcome card on first visit, projects list with status, recent activity from every run.
 </td>
 <td width="50%">
 <a href="docs/assets/screenshots/06-timeline.light.desktop.png">
 <img src="docs/assets/screenshots/06-timeline.light.desktop.png" alt="Live run timeline — planner, coder, reviewer agent cards with token counts and costs." />
 </a>
 Live run timeline — per-agent steps streaming over SSE with input, output, tool calls, token usage and cost as the run unfolds.
 </td>
 </tr>
 <tr>
 <td width="50%">
 <a href="docs/assets/screenshots/07-digest.light.desktop.png">
 <img src="docs/assets/screenshots/07-digest.light.desktop.png" alt="Daily digest — one changeset with Promote, Rollback, and Defer buttons." />
 </a>
 Digest — what the agents produced today, ready for one human decision: Promote, Defer, or Rollback.
 </td>
 <td width="50%">
 <a href="docs/assets/screenshots/10-lifecycle.light.desktop.png">
 <img src="docs/assets/screenshots/10-lifecycle.light.desktop.png" alt="Lifecycle configuration — Coder, Planner, and Reviewer agents with bound skills." />
 </a>
 Lifecycle — agents, workflows, skills, and human gates. Versioned <code>mergecrew.yaml</code>, editable from the UI or your repo.
 </td>
 </tr>
</table>

More captures live in [`docs/assets/screenshots/`](docs/assets/screenshots/). Regenerate with `pnpm screenshots`.

### What you get on first boot

- **Demo mode** is the default — every agent step routes through a deterministic stub, so the full multi-agent loop works with zero LLM keys configured.
- **Welcome card with a one-click sample run** — click the primary CTA on the Today page and you land on the live timeline as agents work.
- **Onboarding wizard** at `/orgs/{slug}/onboarding` walks you through provider → project → repo → deploy target → lifecycle in five DB-derived checklist steps.
- **Stock lifecycle templates** — pick `generic-careful`, `nextjs-vercel`, `python-render`, or `go-fly` from the project Lifecycle page; no YAML editing required.

---

## Why Mergecrew exists

The autonomous-coding category has converged on the *agent layer*: Devin, Cursor's cloud agents, GitHub Copilot's coding agent, OpenHands, SWE-agent. They all do roughly the same thing — take a ticket, open a PR. None of them own the *loop* around that work.

**Mergecrew is the loop.**

- **Scheduled, not triggered.** Runs on a cron against your real repo. No tickets required.
- **Full lifecycle, not just code.** Spec → design → build → deploy-to-dev → bug scan → digest → human approval → prod.
- **Multi-tenant by design.** One workspace per org, tenant-isolated Postgres (RLS), per-org budgets, per-org provider keys, [per-org runner profile (BYO)](docs/03-infrastructure/34-runner-agent.md) so a hosted deployment never runs other tenants' containers on the operator's VM. Self-host privately or run as a service.
- **Human-gated production.** Promotion to prod is always a human decision. Non-configurable invariant.
- **Pluggable agents and providers.** Anthropic, OpenAI, AWS Bedrock, or local Ollama. The bundled NestJS+LangGraph runner ships by default; OpenHands / Claude Code / your own runner can be wired in.
- **Multi-agent specialization.** Optional planner → coder → reviewer split with a reviewer loop-back; cheaper reviewer catches obvious mistakes before a human gets paged. See [`docs/03-infrastructure/18-multi-agent.md`](docs/03-infrastructure/18-multi-agent.md).

## How Mergecrew compares

| Project | Triggered by | Lifecycle scope | Multi-tenant | Prod gate | License |
|---|---|---|---|---|---|
| **Mergecrew** | scheduled (cron) | spec → deploy → digest | yes | mandatory | Apache 2.0 |
| [OpenHands](https://github.com/All-Hands-AI/OpenHands) | task / chat | code only | yes (Cloud) | optional | MIT |
| [Aeon](https://github.com/aaronjmars/aeon) | scheduled (cron) | code only | no | none ("no babysitting") | MIT |
| [SWE-agent](https://github.com/SWE-agent/SWE-agent) | issue | code only | no | n/a | MIT |
| [MetaGPT](https://github.com/FoundationAgents/MetaGPT) / [GPT Pilot](https://github.com/Pythagora-io/gpt-pilot) | one-shot prompt | greenfield only | no | n/a | MIT |
| Devin / Cursor / Copilot agent | task / chat | code only | no | optional | proprietary |

The combination of *scheduled lifecycle + multi-tenant + mandatory prod gate* is currently whitespace. Detailed positioning is in [docs/00-product/02-scope.md](docs/00-product/02-scope.md).

## Architecture in one paragraph

Mergecrew is a pnpm-workspace monorepo. Four NestJS services (`api`, `orchestrator`, `runner`, `worker-cron`) plus a Next.js web app share a Postgres database with row-level security, a Redis instance for BullMQ queues, and an S3-compatible blob store for artifacts. Agents run as [LangGraph](https://github.com/langchain-ai/langgraphjs) state graphs over a pluggable LLM router. Deploy and VCS are behind adapter interfaces — GitHub Actions and Vercel ship by default. The full spec lives in [`docs/`](docs/README.md).

## Status

**Alpha.** The user-facing surfaces (org/project setup, run timeline, digest, approval) are wired end-to-end and run locally against Ollama with no API keys. APIs are not stable; the database schema may change without a migration path between alpha versions. Not yet recommended for production tenants.

---

## Stack

- **Backend.** NestJS 11 across four services: `api`, `orchestrator`, `runner`, `worker-cron`.
- **Frontend.** Next.js 16 (App Router), Tailwind, lucide-react.
- **Data.** Postgres 16 (pgvector + RLS) and Redis (BullMQ + pubsub).
- **Agent runtime.** [LangGraph.js](https://github.com/langchain-ai/langgraphjs) graph + [LangChain.js](https://github.com/langchain-ai/langchainjs) provider integrations (Anthropic, OpenAI, AWS Bedrock, Ollama).
- **Tooling.** pnpm workspaces, Turbo, TypeScript, Prisma.

## Layout

```
apps/
  api/                 NestJS HTTP + SSE                (port 4000)
  orchestrator/        Durable run engine (BullMQ)
  runner/              Agent step executor
  worker-cron/         Schedule scanner
  web/                 Next.js app                       (port 3000)
packages/
  domain/              Zod-shaped domain types
  llm/                 LangChain provider factory + capability router
  skills/              Skill SDK + ~25 stock skills
  agent-runtime/       LangGraph StateGraph (agent + tools nodes)
  adapters-vcs/        GitHub
  adapters-deploy/     GitHub Actions, Vercel
  adapters-tracker/    Linear, GitHub Issues
  adapters-comms/      Slack, Email
  config-yaml/         mergecrew.yaml parser
  db/                  Prisma schema, migrations, RLS
  eventlog/            Timeline events
infra/
  sql/                 RLS + bootstrap
  docker/              Service Dockerfiles
docs/                  Spec set (read this first)
```

---

## Quick start

> **First time?** Follow [`docs/00-quickstart.md`](docs/00-quickstart.md) for a 15-minute step-by-step walkthrough that takes you from `git clone` to your first agent-generated PR using only Docker. The two paths below are the abbreviated versions.

Two paths, pick whichever fits. New here? Pair either path with the
[**mergecrew-sample-app**](https://github.com/mergecrew/mergecrew-sample-app)
— a tiny Next.js app with one deliberate bug, designed to be connected to a
fresh Mergecrew install so you can watch the full agent loop fix it and ship
a Vercel preview, without touching your own codebase.

Pointing Mergecrew at your own repo? Read the
[**deploy-target cookbook**](docs/03-infrastructure/06-deploy-targets-cookbook.md)
first — it has copy-paste configs for the common shapes (Vercel preview,
Netlify, push-triggered ECS Fargate with manual prod gate, raw AWS SDK,
Fly/Render/Railway) so you don't have to assemble one from the adapter
source.

### One command (trial, all-in-Docker)

If you just want to **see Mergecrew running**, the fastest path is the bundled compose stack — Postgres, Redis, API, orchestrator, runner, worker-cron, and web all start together. Migrations apply on boot and a `demo` org + user are seeded so the web UI is usable immediately.

```bash
git clone https://github.com/mergecrew/mergecrew.git
cd mergecrew
pnpm compose:full     # or: docker compose -f docker-compose.full.yml up
```

Open <http://localhost:3000>. Auto-login signs you in as `demo@mergecrew.local`. The bundled demo project ships in **demo mode** (`MERGECREW_DEMO_MODE=1`), so you can click **Run now** immediately — the agent loop returns a canned plan/diff/verdict instead of calling an LLM, and an amber banner makes the mode unambiguous. Set `MERGECREW_DEMO_MODE=0` and add an LLM provider in the org settings once you want real runs. Stop with `pnpm compose:full:down`; wipe state with `docker compose -f docker-compose.full.yml down -v`.

This path uses dev-mode defaults (zero JWT secret, no OAuth, no TLS) and is for local trials only. Production self-host uses `docker-compose.prod.yml` — see [docs/03-infrastructure](docs/03-infrastructure). For an automated single-VM deploy via GitHub Container Registry, see [docs/03-infrastructure/19-vm-deploy.md](docs/03-infrastructure/19-vm-deploy.md).

### Hacking on the code (recommended for contributors)

You'll be running everything locally with **Docker for Postgres/Redis**, **Homebrew Ollama for a local LLM**, and **Node + pnpm for the apps**. No paid services are required, no OAuth setup is required to log in.

### 1. Prereqs

| Tool | Purpose | Install |
|---|---|---|
| Node 22+ (LTS) | runtime | `nvm install --lts && nvm alias default lts/*` |
| pnpm 9+ | package manager | `npm install -g pnpm@9.12.0` |
| Docker | Postgres + Redis | Docker Desktop |
| Homebrew | macOS package manager (used for Ollama) | https://brew.sh |
| Ollama (optional but recommended) | local LLM, no API keys | `brew install ollama` |

The repo's `package.json` pins `pnpm@9.12.0` via `packageManager`, so Corepack will use that exact version automatically.

### 2. Clone and install

```bash
git clone https://github.com/mergecrew/mergecrew.git
cd mergecrew
cp .env.example .env
pnpm install
```

The default `.env` works out of the box for local dev. **Auto-login is enabled by default** (`MERGECREW_DEV_AUTO_LOGIN=true`), so you do not need to set up GitHub or Google OAuth to use the app — you'll be signed in as a built-in demo user.

### 3. Start Postgres + Redis (Docker)

```bash
pnpm compose:up
```

This brings up `mergecrew-postgres` (pgvector/pg16) on `:5432` and `mergecrew-redis` on `:6379`. The Postgres init script creates the `mergecrew`, `mergecrew_app`, and `mergecrew_migrator` roles, the `mergecrew` database, and the `pgvector` / `uuid-ossp` / `pgcrypto` extensions.

To stop: `pnpm compose:down`. To wipe: `docker compose down -v`.

### 4. Apply migrations and seed

```bash
pnpm db:migrate
pnpm db:seed
```

The seed creates:
- Organization `demo` (slug)
- User `demo@mergecrew.local` with role `owner`
- Project `demo-saas` with a default lifecycle (`mergecrew.yaml`) and ~8 agents — flagged `demo=true` (read-only outside `MERGECREW_DEMO_MODE`)

> The `pnpm db:*` and `pnpm dev` scripts auto-load `.env` via `dotenv-cli`. If you invoke `prisma` directly outside pnpm, prefix with `pnpm exec dotenv -e .env -- prisma …` so the connection URLs are visible.

### 5. Start a local LLM (Ollama)

You can skip this if you intend to use a hosted provider (Anthropic, OpenAI, Bedrock). For the zero-cost path:

```bash
brew services start ollama
ollama pull qwen3:0.6b      # ~520MB, supports tool use
```

Verify Ollama is up:

```bash
curl http://localhost:11434/api/tags
```

### 6. Run the app

For development with hot reload across all four backend services + the web app:

```bash
pnpm dev
```

This runs `tsx watch` against `api`, `orchestrator`, `runner`, `worker-cron`, plus `next dev` for `web`, all in parallel.

For a production-equivalent local run:

```bash
pnpm -r build
pnpm --filter @mergecrew/api start          &
pnpm --filter @mergecrew/orchestrator start &
pnpm --filter @mergecrew/runner start       &
pnpm --filter @mergecrew/web start          &
```

### 7. Open the app

Browse to <http://localhost:3000>. You'll land directly on **Demo Org** — no sign-in required.

### 8. Register the LLM provider and trigger a run

Out of the box, no LLM provider is registered for the demo org — you have to point Mergecrew at one. The fastest path is the local Ollama provider:

```bash
USER_ID=$(PGPASSWORD=mergecrew psql -h localhost -U mergecrew -d mergecrew -tA -c \
  "select id from users where email='demo@mergecrew.local';")
H="x-mergecrew-user-id: $USER_ID"

# 1. Register the Ollama provider
PID=$(curl -sS -X POST http://localhost:4000/v1/orgs/demo/llm/providers \
  -H "$H" -H 'content-type: application/json' \
  -d '{"kind":"ollama","label":"Local Ollama","endpoint":"http://localhost:11434",
       "capabilityOverrides":{"models":["qwen3:0.6b"]}}' | jq -r .id)

# 2. Default profile that prefers qwen3:0.6b
curl -sS -X POST http://localhost:4000/v1/orgs/demo/llm/profiles \
  -H "$H" -H 'content-type: application/json' \
  -d "{\"name\":\"local\",\"preferenceOrder\":[\"$PID/qwen3:0.6b\"],\"capabilityRouting\":{}}"

# 3. Trigger a daily run
curl -sS -X POST http://localhost:4000/v1/orgs/demo/projects/demo-saas/runs -H "$H" -d '{}'
```

The run takes ~50 seconds on Ollama qwen3:0.6b. Watch it progress at <http://localhost:3000/orgs/demo/projects/demo-saas/timeline>.

To inspect run state directly:

```bash
psql "postgresql://mergecrew:mergecrew@localhost:5432/mergecrew" -c \
  "select agent_kind, status, total_input_tokens, total_output_tokens
   from agent_steps order by started_at desc limit 8;"
```

---

## Environments and ports

| Service | URL | Notes |
|---|---|---|
| Web | <http://localhost:3000> | Next.js |
| API | <http://localhost:4000> | NestJS HTTP + SSE |
| OpenAPI | <http://localhost:4000/v1/openapi.json> | |
| Postgres | localhost:5432 | user `mergecrew` / pw `mergecrew` / db `mergecrew` |
| Redis | localhost:6379 | |
| Ollama | localhost:11434 | only if installed |

Logs are streamed to `.logs/{api,orchestrator,runner,web}.log` if you started the services in the background.

## Authentication

### Default: dev auto-login

The web app is signed in as **`demo@mergecrew.local`** automatically. This is controlled by:

```env
MERGECREW_DEV_AUTO_LOGIN=true            # default in non-production
MERGECREW_DEV_USER_EMAIL=demo@mergecrew.local
MERGECREW_DEV_USER_NAME=Demo User
BFF_TRUST_TOKEN=dev-trust-token        # the API trusts the web BFF when this matches
```

When `NODE_ENV=production`, auto-login defaults to off. Set `MERGECREW_DEV_AUTO_LOGIN=false` explicitly to disable it in dev too.

The API also accepts an `x-mergecrew-user-id` header for programmatic access (used by the curl examples above) — also a dev-only convenience.

### Optional: GitHub / Google OAuth

If you want real OAuth in your local environment:

1. Set `MERGECREW_DEV_AUTO_LOGIN=false` in `.env`.
2. Configure sign-in providers:
 - GitHub: reuses the GitHub App you set up for repo integration. In the App's settings → **Identifying and authorizing users**, set the callback URL to `http://localhost:3000/api/auth/callback/github` (or your prod equivalent). NextAuth reads `GITHUB_APP_CLIENT_ID` / `GITHUB_APP_CLIENT_SECRET` from `.env`.
 - Google (optional): <https://console.cloud.google.com/apis/credentials> — callback `http://localhost:3000/api/auth/callback/google`. Fill `GOOGLE_OAUTH_CLIENT_ID` / `GOOGLE_OAUTH_CLIENT_SECRET`.
3. Restart the web app.

## Switching to a hosted LLM (Anthropic / OpenAI / Bedrock)

The provider abstraction routes through LangChain's official integrations, so no custom code is required to switch:

```bash
# Anthropic example — needs ANTHROPIC_API_KEY in .env first
PID=$(curl -sS -X POST http://localhost:4000/v1/orgs/demo/llm/providers \
  -H "$H" -H 'content-type: application/json' \
  -d '{"kind":"anthropic","label":"Anthropic","apiKey":"'"$ANTHROPIC_API_KEY"'",
       "capabilityOverrides":{"models":["claude-sonnet-4-6"]}}' | jq -r .id)

curl -sS -X POST http://localhost:4000/v1/orgs/demo/llm/profiles \
  -H "$H" -H 'content-type: application/json' \
  -d "{\"name\":\"anthropic\",\"preferenceOrder\":[\"$PID/claude-sonnet-4-6\"],\"capabilityRouting\":{}}"
```

Credentials are envelope-encrypted at rest using `KMS_MASTER_KEY` from `.env`.

## Common scripts

| Command | What it does |
|---|---|
| `pnpm install` | install all workspaces |
| `pnpm compose:up` / `pnpm compose:down` | Postgres + Redis (Docker) |
| `pnpm db:migrate` | apply migrations |
| `pnpm db:seed` | seed demo org + project |
| `pnpm db:reset` | drop + recreate + migrate + seed |
| `pnpm dev` | run all services in watch mode |
| `pnpm build` | build all packages and apps |
| `pnpm typecheck` | typecheck across the workspace |
| `pnpm test` | run tests |

## Troubleshooting

- **Web shows "create your first organization"** — the seed didn't run (or ran against a different DB). `pnpm db:seed`.
- **API returns `org not found`** — your `x-mergecrew-user-id` doesn't match the seeded user. Look it up:
  `psql "$DATABASE_URL" -c "select id, email from users;"`
- **Run fails with `no provider available`** — register an LLM provider for the org (step 8 above).
- **Ollama model not found** — `ollama pull qwen3:0.6b`.
- **`prisma migrate deploy` complains about `DATABASE_MIGRATE_URL`** — you ran Prisma directly. Use `pnpm db:migrate` (which auto-loads `.env`), or prefix with `pnpm exec dotenv -e .env -- prisma migrate deploy`.
- **Port 5432 / 6379 already in use** — you have a local Postgres/Redis already running. Either stop it or change the port mapping in `docker-compose.yml`.

---

## Why this stack

- **LangChain.js + LangGraph.js** for the agent loop — battle-tested OSS, handles provider drift, gives you a graph the team already knows.
- **BullMQ + Postgres** for the workflow layer — durable queues survive restarts and pause-on-rate-limit; LangGraph isn't designed to replace this.
- **NestJS + Next.js** because most TS engineers can read both immediately, lowering the contributor ramp-up.
- **No paid runtime dependencies.** Observability hooks are open ([Langfuse](https://langfuse.com) drops in via LangChain callbacks) but not required.

## Contributing

Mergecrew is built in the open and accepts contributions. See [CONTRIBUTING.md](CONTRIBUTING.md) for the dev setup and PR flow. Start with the [`docs/`](docs/README.md) tree to understand the architecture and product invariants, then check [open issues](https://github.com/mergecrew/mergecrew/issues) for `good first issue` or `roadmap`. PRs that change agent behavior, multi-tenant isolation, or the human approval gate require a design note in the PR description and ideally a [Discussion](https://github.com/mergecrew/mergecrew/discussions) first.

By participating you agree to the [Code of Conduct](CODE_OF_CONDUCT.md). Security issues should be reported privately per [SECURITY.md](SECURITY.md).

## Sponsors

Mergecrew runs autonomous LLM workflows daily; even alpha-stage dogfooding burns real money in API credits. The project is looking for backers — Anthropic / OpenAI / Bedrock credits, hosting credits, or recurring sponsorship via GitHub Sponsors. See [SPONSORS.md](SPONSORS.md) for what's needed and how to back the project.

> **Looking for the first sponsor.** Token-credit and cash sponsors are acknowledged in `SPONSORS.md`, in release notes, and on this README.

## License

[Apache 2.0](LICENSE).

mergecrew

파일 탐색기

# CDN으로 사용하기

Was this content helpful?

Recent Feedback

Download README