Skip to content

Getting started

What you need

  • Node 22.18 or later. Keel runs the TypeScript file directly, with no build step. The repository includes an .nvmrc, so nvm use picks the right version.
  • A model to talk to. Any server that speaks the OpenAI-style /chat/completions API works. The quickest option is Ollama, which runs models on your own machine. Models covers Ollama, hosted APIs and AWS Bedrock.

Run Keel

Pull a model, clone the repository, move into the folder where you want Keel to work, and run it:

ollama pull qwen3-coder:30b
git clone https://github.com/Rayjk123/Keel.git
mkdir scratch && cd scratch

BASE_URL=http://localhost:11434/v1 MODEL=qwen3-coder:30b \
  node ../Keel/agent.ts "create a script that prints the first 20 primes"

Keel works in the folder you run it from and runs real commands there, so a scratch directory is a good place to start.

As it works, Keel prints each command it runs, prefixed with $. When the model is finished, it prints a short summary of what it did.

Settings

Keel is configured with environment variables.

Setting Default What it does
BASE_URL http://localhost:8000/v1 The chat API to call
MODEL required The model name
API_KEY none Bearer token, if your API needs one