docs: add official Ollama/llama.cpp/vLLM local model examples

[Yeachan-Heo]

Source

Discord #claw-code support: users asked whether Claw Code can run all-local with Ollama/local Qwen/llama.cpp and how to configure a non-Anthropic provider.

Problem

Community guides exist, but local/OpenAI-compatible setup should have official known-good examples and warnings.

Suggested scope

Add docs for local OpenAI-compatible providers:

• Ollama
• llama.cpp server
• vLLM or other OpenAI-compatible servers

Include examples such as:

export OPENAI_BASE_URL="http://127.0.0.1:11434/v1"
export OPENAI_API_KEY="dummy"
claw --model "openai/qwen3:latest" prompt "Reply exactly HELLO_WORLD_123"

Explain:

• use openai/ prefix for OpenAI-compatible route
• model id must match what the local server exposes
• raw /v1/chat/completions smoke test
• tool-call support may be required for slash/tool workflows
• local routes may be rougher than first-party Anthropic/OpenAI

Acceptance criteria

• Users can configure at least one local provider from official docs.
• Docs include a one-shot prompt smoke test.
• Docs warn about tool-call/response-shape limitations.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]