API Reference

Endpoints implemented by the standalone proxy server.

LLM endpoints

  • POST /v1/messages — Anthropic Messages API passthrough/routing (streaming + non-streaming)
  • POST /v1/messages/count_tokens — Anthropic token counting passthrough
  • POST /v1/chat/completions — OpenAI-compatible chat completions routing
  • GET /v1/models — returns RelayPlane virtual models

Implementation note: model list and chat completions use URL substring matching (/models, /chat/completions).

Health & status

  • GET /status
  • GET /health
  • GET /healthz (alias of /health)

Control endpoints

  • POST /control/enable
  • POST /control/disable
  • GET /control/status
  • GET /control/stats
  • POST /control/config (JSON merge patch)
  • POST /control/kill (body: {"sessionKey":"..."} or {"all":true})

/control/* endpoints are administrative. Keep them bound to 127.0.0.1 by default, and only expose them with strong network/auth controls.

Telemetry endpoints

  • GET /v1/telemetry/stats (query: ?days=7)
  • GET /v1/telemetry/runs (query: ?limit=50&offset=0)
  • GET /v1/telemetry/savings
  • GET /v1/telemetry/health

Mesh + config + dashboard

  • GET /v1/mesh/stats
  • POST /v1/mesh/sync
  • GET /v1/config
  • GET / and GET /dashboard
  • GET /dashboard/config

Security notes

Treat /control/*, /v1/config, and /dashboard/config as sensitive admin endpoints. Keep the proxy bound to 127.0.0.1 by default. If you must bind to 0.0.0.0, require a firewall, reverse proxy controls, and strong authentication before exposing it beyond localhost.

Example request

1curl -X POST http://localhost:4100/v1/chat/completions \
2  -H "Content-Type: application/json" \
3  -d '{
4    "model": "relayplane:auto",
5    "messages": [{"role":"user","content":"hello"}]
6  }'
Unknown routes return 404. The fallback error currently lists supported core endpoints: POST /v1/messages, POST /v1/chat/completions, GET /v1/models.