Using Atomic Chat With Kilo Code

Kilo Code supports Atomic Chat as a local provider. Atomic Chat runs models on your machine and exposes an OpenAI-compatible API (default http://127.0.0.1:1337/v1).

Website: https://atomic.chat/
Repository: https://github.com/AtomicBot-ai/Atomic-Chat

Prerequisites

  1. Install Atomic Chat (macOS or Windows).
  2. Download and load a model in the app.
  3. Enable the local API server (default port 1337).
  4. Confirm the API responds:
curl http://127.0.0.1:1337/v1/models

Configuration in Kilo Code

Kilo Code ships the @kilocode/plugin-atomic-chat plugin by default. It does not call localhost unless you opt in (see below). When enabled, it discovers models from GET /v1/models and can warn if the selected model is not loaded.

Localhost HTTP runs only when one of these is true:

  • You configure provider.atomic-chat in kilo.jsonc
  • You set "model": "atomic-chat/..." (or per-agent model uses atomic-chat)
  • You enable optional auto-detect: "atomicChat": { "autoDetect": true } (probes ports 1337 and 1338)

Otherwise no requests are made to Atomic Chat (suitable for restricted environments).

Open Settings (gear icon) → Providers → Atomic Chat. No API key is required for the default local server. Adjust the base URL if Atomic Chat uses a non-default host or port.

Custom or unlisted models

If a loaded model does not appear in the picker, register it under provider.atomic-chat.models:

{
  "model": "atomic-chat/my-local-model",
  "provider": {
    "atomic-chat": {
      "models": {
        "my-local-model": {
          "id": "exact-id-from-v1-models",
          "name": "My Local Model",
        },
      },
    },
  },
}

See Custom Models for all model fields.

Tips

  • Prefer capable models with large context windows; agent workflows use long prompts.
  • Keep only the models you need loaded in Atomic Chat to save memory.
  • For embeddings via Atomic Chat, use the openai-compatible indexing provider with the same base URL.