Models
Keel talks to any server that speaks the OpenAI-style /chat/completions API. Pick where your model runs, then point Keel at it with BASE_URL, MODEL and, if the API needs one, API_KEY.
Ollama
Ollama runs models on your own machine:
ollama pull qwen3-coder:30b
BASE_URL=http://localhost:11434/v1 MODEL=qwen3-coder:30b node agent.ts "your task"
A hosted API
Point BASE_URL at any OpenAI-compatible endpoint and set MODEL and API_KEY:
BASE_URL=https://api.example.com/v1 MODEL=my-model API_KEY=sk-... \
node agent.ts "your task"
AWS Bedrock
Bedrock has its own API, so Keel reaches it through a small local proxy in bedrock/ that signs each request with your AWS credentials. AWS Bedrock covers setting it up and using it in benchmarks.