Production
Deployment
Run the Antraft control plane as a service — it exposes the governance endpoint, fleet control, the observability dashboard, and the approval inbox.
Run the control plane
shell
pip install "antraft[all]" "uvicorn[standard]"antraft serve --host 0.0.0.0 --port 8000Endpoints
| Route | Purpose |
|---|---|
POST /guard/evaluate | Language-agnostic governance (used by the TS SDK). |
GET /agents | List active agents. |
POST /agents/{id}/approve | Approve a paused action. |
POST /agents/{id}/signal | PAUSE / RESUME / STOP. |
GET /observability | Live governance dashboard. |
GET /inbox | Human-in-the-loop approval inbox. |
Docker
shell
docker build -t antraft .docker run -p 8000:8000 antraftMCP server
Expose Antraft itself as an MCP server for Claude Desktop or an IDE:
shell
antraft mcp # stdio MCP serverIn-process vs sidecar
For Python apps, the runtime governs in-process — fast and embedded. For other languages (or a central control plane), run antraft serve as a sidecar and call/guard/evaluate. Both paths use the same engine, so decisions never diverge.