1. Install Kimi Code
Install with the standalone installer:2. Add Together AI as a provider
Kimi Code reads providers and models from~/.kimi-code/config.toml. Add Together AI as an OpenAI-compatible provider, register Kimi K3 under it, and make it the default model:
~/.kimi-code/config.toml
your_together_api_key with your actual Together AI API key.
If the file already exists, keep any settings you have and add the default_model line, the [providers.together] table, and the [models."together/kimi-k3"] table alongside them.
Validate the configuration:
3. Run Kimi Code
Launch the agent from your project directory:/model inside the session to confirm the model or switch between any models you have configured. Kimi K3 thinks by default, and Kimi Code renders the reasoning trace as the agent works.
Alternative: configure with environment variables
For a quick test without touchingconfig.toml, you can select the model entirely through the KIMI_MODEL_* environment variables. Kimi Code synthesizes a temporary provider from them for that session only:
config.toml setup above for day-to-day work.
That’s it! You now have Moonshot’s own coding agent running Kimi K3 on Together AI. See the Kimi K3 quickstart for model details and pricing, and the Kimi Code documentation for the agent’s full feature set.