agentic-dev-team is a self-hosted autonomous coding team: you describe what to build and approve ideas, and a set of AI agents handle planning, coding, review, testing, and security across the whole pipeline. A shared work board is the coordination backbone — agents react to events rather than talking to each other directly.
Key Features
- Stack-aware workflows (Python, Node + TypeScript, Go, and more) and SDLC styles (standard, TDD, spec-first)
- Quality gates: stack-appropriate CI, in-coder TDD before every PR, merge-conflict recovery, and post-merge CI validation
- Per-role model routing — frontier models for planning, cheaper models for high-volume execution
- Cost telemetry with daily spend caps
- Single-operator security model (HTTP Basic Auth), designed for a laptop or trusted LAN
- Agent roles: Idea, Planner, Coder, Reviewer, Tester, and Security
Tech: FastAPI + SQLite + Redis (event bus and board), Forgejo for self-hosted git, and Docker Compose. Agents are built on OpenCode + LiteLLM and are model-agnostic — route through OpenRouter, the Anthropic/Claude API, or a local Ollama instance.
This is the tool I used to build k5n-mcp-hub — a dedicated write-up is coming.