Models & Modes

Modes define behavior; models define the underlying AI. In most cases you pick a Mode and let Carrier OS choose the model — but you can override the model when you need to.

Carrier Modes — behavior

A Carrier Mode is a stable, provider-agnostic behavior preset. The CLI fetches the live list from the Models API (GET /v1/models); commonly available modes include:

  • general — everyday questions, writing, and summarization.
  • code — coding, debugging, and technical problem-solving.
  • research — in-depth analysis and long-form synthesis.

Set or list modes with /mode and /modes. Run /modes to see exactly what your project exposes — see also Carrier Modes.

Provider models — the underlying AI

A provider model is the specific model behind a turn. The CLI fetches the selectable list from GET /v1/models/chat — for example GPT, Claude, Gemini, Grok, and Llama family models. Choose one with /model, or select Auto to let the current mode pick.

text
❯ /model            # arrow-key picker (Auto + each model)
❯ /model auto       # let the mode decide
❯ /models           # list provider models

Modes vs models — the rule of thumb

  • Change Mode to change how Carrier responds (its behavior).
  • Change Model to change which AI answers (the engine).
Modes keep your setup future-proof: when the model behind a mode is upgraded, your workflow keeps working unchanged. Both selections apply to your next message and persist for the session.