Skip to content

Auxiliary Models

Not every job needs your main model. ChatShell supports two auxiliary model roles that you can configure separately in Settings → Auxiliary Models.

Used for small background jobs:

  • Conversation titles — automatic title generation
  • Compaction summaries — the text summary written when a long conversation is compacted (see Context Compaction)

By default the tiny role follows your conversation’s model. Point it at a cheap, fast model (for example a small local model or a low-cost cloud model) and titles/summaries stop billing your main model.

Used for image understanding:

  • Image questions — when the AI uses the Read tool on an image with a question, the answer comes from the vision model
  • Images with text-only main models — keep attaching images even when your main model cannot see; ChatShell annotates them so the AI can still inspect them through the vision model

Defaults: the vision role follows your conversation’s model when it is vision-capable. When the main model has no vision, the role is left unset — configure one explicitly to unlock image features. ChatShell warns in the input box when the active model cannot process images and no vision fallback exists.

  • Cost — titles and compaction summaries are high-frequency, low-difficulty work; a tiny model does them for a fraction of the price.
  • Freedom of choice — pair a strong text-only coder as your main model with a cheap vision model for images, without giving up either capability.
  • Graceful degradation — if an auxiliary model is missing or a call fails, the feature degrades silently to the previous behavior; nothing breaks mid-conversation.