Besides the hosted models and your own coding-agent subscriptions through Kai Code Bridge, Kai Code can run on models you host yourself. Local models are configured under AI agents → Kai Code → Settings → Local LLM and appear in the composer's model picker with a Local label. Turns on a local model run on your machine and are not billed by Gleap.
Ollama
Models from a local Ollama install (default http://127.0.0.1:11434). Pull a model on the machine and open the tab. Kai detects installed models automatically, or use Detect. Added models become selectable in Kai Code's local runtime.
Pick a model that supports tool calls and fits your memory. Coding sessions rely heavily on tools (reading files, running commands), so a model without tool support cannot complete a task.
Custom OpenAI-compatible endpoints
Any server that speaks the OpenAI chat-completions API: llama.cpp, LM Studio, vLLM, or a box on your LAN. Choose Add endpoint and enter:
Label: the slug shown in the model picker. Models appear as
<label>/<model>.Base URL: the OpenAI-compatible root of the server, including
/v1.API key: optional; leave empty for local servers.
Model id and Context window for each model the endpoint serves.