Skip to content

openclaw.md

Latest commit

 

History

History
32 lines (20 loc) · 1.1 KB

openclaw.md

File metadata and controls

32 lines (20 loc) · 1.1 KB

OpenClaw

OpenClaw is an open coding agent that targets local models via the openai-completions API mode.

1. Generate the provider config

afm prints a paste-ready config for you:

afm mlx -m mlx-community/Qwen3-Coder-Next-4bit --openclaw-config

This emits a JSON block with baseUrl, api: "openai-completions", model metadata (vision / reasoning detection, context window, max tokens), and zero-cost pricing fields. Copy it into your OpenClaw provider config.

2. Start the server

afm mlx -m mlx-community/Qwen3-Coder-Next-4bit \
  --port 9999 --enable-prefix-caching

(--openclaw-config defaults to port 9999 unless you pass -p.)

3. Run OpenClaw

OpenClaw will issue requests against http://localhost:9999/v1. Tool calling, streaming, and reasoning extraction all work out of the box.

Notes

  • The generated config detects <think> reasoning support automatically.
  • Vision capability is reported as true for VLM-class model IDs.
  • For multi-session use, add --concurrent N (each OpenClaw conversation can hold its own slot).