🌐 Language: English | Русский | 中文

Multi-agent software development platform powered by LangGraph

AI-crew orchestrates teams of AI agents that build software autonomously — from discussing implementation details with the user to deploying the finished project and delivering a live URL. The platform ships a growing collection of agent team graphs tailored for different scenarios: full-cycle development teams, lightweight coding assistants, and research crews.

1. You describe the task — via Web UI or Telegram bot
2. AI manager discusses the plan — clarifies requirements, proposes architecture, agrees on details
3. Agent team executes — analysts, architects, developers, reviewers, QA work together autonomously
4. You watch it happen — interactive graph visualization shows every agent step in real time
5. You get the result — deployed project with a live URL delivered to you

• Multiple agent team configurations — pick the right team for the job, from a solo developer to a full 7-agent crew
• End-to-end delivery — the cycle doesn't stop at a PR; the project gets deployed and you receive a working URL
• Human-in-the-Loop — AI manager discusses implementation details with you before the team starts building
• Interactive graph visualization — watch every agent node execute in real time on a live visual graph
• Telegram integration — create and manage tasks directly from Telegram
• Escalation ladder — automatic escalation when Dev↔QA cycles get stuck
• Observability — full tracing and debugging via Langfuse
• Docker ready — dev (docker-compose) and prod (all-in-one image)

  Telegram ─────┐
                ▼
  Web UI ──► Gateway API ──► LangGraph Engine
                                    │
          ┌─────────────────────────┤
          ▼                         ▼
   ┌─ dev_team ──────┐     ┌─ research ──────┐
   │ PM → Analyst →  │     │ Researcher →    │
   │ Architect →     │     │ Web Search →    │
   │ Developer →     │     │ Report          │
   │ Security →      │     └─────────────────┘
   │ Reviewer → QA   │
   └──────┬──────────┘     ┌─ simple_dev ────┐
          │                │ Developer →     │
          ▼                │ Commit          │
   CI/CD → Deploy          └─────────────────┘
          │
          ▼
   Live URL → User

   PostgreSQL  │  Langfuse  │  GitHub

# 1. Set up environment
cp env.example .env
# Fill in LLM_API_KEY in .env

# 2. Start all services
docker-compose up -d

# 3. Start the frontend
cd frontend && npm install && npm run dev

Open http://localhost:5173, enter a task and watch the agents work on the interactive graph.

More details: docs/GETTING_STARTED.md

AI-crew/
├── graphs/                   # Agent team graphs
│   ├── dev_team/             #   Full 7-agent development team
│   ├── standard_dev/         #   Medium-complexity development
│   ├── simple_dev/           #   Fast single-agent coding
│   ├── research/             #   Research & analysis
│   └── common/               #   Shared utilities, types, git, logging
├── frontend/                 # React Web UI with graph visualization
├── gateway/                  # API gateway (FastAPI)
├── telegram/                 # Telegram bot
├── tests/                    # Tests (pytest)
├── vendor/aegra/             # Aegra server (vendored)
├── scripts/                  # Docker entrypoint, setup, nginx
├── docs/                     # Documentation
├── docker-compose.yml        # Development
├── Dockerfile                # Production (all-in-one)
├── aegra.json                # Aegra config
└── env.example               # .env template

pip install -r requirements.txt
pytest tests/ -v

• Prompts — graphs/*/prompts/*.yaml
• Models — env LLM_MODEL_PM, LLM_MODEL_DEVELOPER, etc.
• New agent — see docs/DEVELOPMENT.md
• New graph — add a directory under graphs/ with graph.py and manifest.yaml

MIT