Skip to content

SERVER.md

Latest commit

 

History

History
86 lines (63 loc) · 1.58 KB

SERVER.md

File metadata and controls

86 lines (63 loc) · 1.58 KB

HTTP API Server

enzu includes a FastAPI server for running tasks over HTTP.

Install

uv pip install "enzu[server]"
# or: pip install "enzu[server]"

Configure

export OPENAI_API_KEY=sk-...
# Optional auth (requires X-API-Key on requests):
# export ENZU_API_KEY=your-secret

# Optional defaults:
# export ENZU_DEFAULT_MODEL=gpt-4o
# export ENZU_DEFAULT_PROVIDER=openai

Run

uvicorn enzu.server:app --host 0.0.0.0 --port 8000

Run a task

curl http://localhost:8000/v1/run \
  -H "Content-Type: application/json" \
  -d '{"task":"Say hello","model":"gpt-4o","provider":"openai"}'

Response shape:

{
  "answer": "...",
  "request_id": "...",
  "model": "...",
  "usage": {
    "total_tokens": 123,
    "prompt_tokens": 45,
    "completion_tokens": 78,
    "cost_usd": 0.0123
  }
}

Sessions (multi-turn)

Create a session:

curl http://localhost:8000/v1/sessions \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o","provider":"openai","max_cost_usd":1.0}'

Run inside the session:

curl http://localhost:8000/v1/sessions/<session_id>/run \
  -H "Content-Type: application/json" \
  -d '{"task":"What did I just say?"}'

Fetch session state:

curl http://localhost:8000/v1/sessions/<session_id>

Auth

If ENZU_API_KEY is set, include X-API-Key on every request.

Notes

  • Sessions are in-memory and scoped to a single server process.
  • Use ENZU_DEFAULT_MODEL and ENZU_DEFAULT_PROVIDER to avoid sending model/provider each request.
  • GET /health is available for load balancers.