Harness docs / Get started

Choosing a model

With the default "model": "auto" there is nothing to configure: the harness looks for a local server (llama.cpp on 8081/8080, Ollama on 11434, LM Studio on 1234), then for ANTHROPIC_API_KEY, OPENAI_API_KEY, then OPENROUTER_API_KEY, and uses the first it finds. It prints what it picked. If the servers list several models, the first interactive start shows them numbered (embedding models left out, a model from the target range below marked *), you pick one, and it is saved as chosenModels in settings.local.json, so later runs use it without asking. If that model is no longer listed, it asks again and shows every model it found. Runs that cannot ask (--auto, -p, --check, no terminal) never wait: they use the saved model, or else the best candidate for that run only, and print a warning. To choose yourself, run /model list [text] (every model your local servers and keys offer, numbered), then /model <number>, or name one directly: --model openrouter:anthropic/claude-sonnet-4.5 (provider:model-id, provider = anthropic, openai, openrouter or a profile in settings). Pin a whole profile with --model <name>, /model <name>, or model in the settings.