trace-proxy

Overview ¶

Package main implements the Aleutian Trace OpenAI-compatible proxy.

Description:

This binary provides an OpenAI-compatible API endpoint that translates
chat completion requests into trace agent loop calls. Any OpenAI-compatible
client (Open WebUI, Continue.dev, Cline, Aider) can connect to this proxy
and get full CRS reasoning + trace tools transparently.

The proxy does NOT do its own tool loop. The agent loop handles everything:
tool selection (via CRS), tool execution, multi-step reasoning, and response
synthesis. The proxy is just a protocol translator.

Usage:

trace-proxy [flags]
  --listen-addr string   Listen address (default: ":12218")
  --trace-url string     Trace server URL (default: ALEUTIAN_TRACE_URL env or http://localhost:12217)
  --ollama-url string    Ollama URL for /v1/models (default: OLLAMA_URL env or http://localhost:11434)
  --project-root string  Default project root (default: current directory, can be set per-request via X-Project-Root header)
  --timeout duration     Agent run timeout (default: 5m)

Example:

# Start proxy with default project root
trace-proxy --project-root /path/to/repo

# Test with curl
curl -X POST http://localhost:12218/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "glm4:latest", "messages": [{"role": "user", "content": "What functions call parseConfig?"}]}'

# Point Open WebUI at http://localhost:12218 as the API endpoint

Package main implements the Aleutian Trace OpenAI-compatible proxy.