Skip to main content

Configuring Models

MihaEL uses a main model for chat and tool loops. Helper jobs (titles, compression, vision) can use the same model or a cheaper one.

In the app​

Settings → Model sets the default for new chats, cron, and subagents.

The composer picker (left of the microphone) changes the model for the current chat only. It does not rewrite your default.

  • Switching models mid-chat re-reads the conversation at full price (prompt cache is per model). Prefer a new chat when you change models on a long thread.
  • Each profile has its own default.

Providers​

Settings → Providers:

  • Accounts — OAuth sign-in (Nous Portal, ChatGPT/Codex, Anthropic, and others the app lists)
  • API keys — paste a key for OpenRouter, OpenAI, Gemini, and the rest
  • Custom endpoints — an OpenAI-compatible URL (Ollama, vLLM, llama.cpp, …)
  • Local AirLLM — optional on-device models; start a new session after enabling

Nous Portal is the fastest path: one sign-in covers models plus web search, image generation, TTS, and cloud browser.

Auxiliary slots​

Settings → Model can show extra slots (vision, compression, titles). Leave them on auto unless you want a cheaper model for those jobs.

Optional on-device inference is separate: Local AirLLM.