Choose your default models
Set the model, effort and fallbacks your own conversations start with, without changing what anyone else uses.
Default model
- Open Account → Models.
- Choose a model under Primary model. Pick Inherit to follow the instance default instead.
- Choose another conversation and send a message. The new conversation uses your choice.
Every conversation without a model of its own starts with this one. A conversation keeps the model it was given across reconnects, and switching models inside a conversation with the picker or /model applies to that conversation only.
A new choice reaches a running conversation when it next starts up, so a turn already in progress finishes on the model it began with.
If you switch models while working in a Git project, Chetty remembers the choice for that project and uses it in later conversations there. Clear the project's choice to go back to your primary model.
Note: A model that is unavailable or no longer allowed is replaced by an applicable project, account or server default. Chetty does not report a model it cannot reach.
Reasoning effort
Reasoning effort sets how much work the model does before it answers. The levels offered depend on the selected model, and Default leaves the choice to the model.
Higher effort can take longer and use more tokens. A saved level that the new model does not support is cleared when you change models, so you are never left on a setting the model cannot honour.
Vision model
Vision fallback model handles messages with images when your chat model cannot read them. Leave it set to inherit to use the primary model, and Chetty says so next to the option when that model cannot read images; the image is then not processed.
A turn with an image uses the fallback, and the next text-only turn goes back to your normal model.
Fast mode
Fast mode asks the provider for priority processing, and it is chosen per model, not per account. Only models whose provider supports it are listed:
- ChatGPT sign-in, OpenAI and Azure OpenAI API keys, and Anthropic API keys.
- For the models the catalog marks as supporting Fast.
- Never models on a Claude subscription.
The list follows the model catalog, so a newly added model appears once the catalog is refreshed. A model you did not select never uses Fast.
The status line shows FAST only while the current conversation's model has it on. /fast switches it for the current conversation's model alone, and says so when that model does not support it. Running conversations pick up a changed selection on their next provider request.
When none of your models qualifies, the row reads that Fast mode is not available for your models and names the connection types that would make it available.
Automatic compaction
When a conversation nears its context limit, Chetty summarizes older context to make room.
- Automatically shorten context is on by default.
- Shorten when context is full sets the threshold in percent. The default is 80%, and accepted values are 30–95%.
- Compaction per model gives selected models their own threshold, because the same percentage is a different amount of text per model.
- Compaction model chooses the model that writes the summaries, including for
/compact.
Leave Compaction model empty and every provider uses the conversation's model, including ChatGPT sign-in, so the summary can reuse the context already cached. An unavailable or revoked pick also falls back to the conversation's model.
Compaction applies to existing conversations as well as new ones. A conversation in a channel keeps automatic compaction on and uses the threshold of the account it composes as. See Chat overview for what compaction keeps and what it drops.
If a model disappears
A provider that is disconnected, a subscription that ends or a model removed from the catalog can empty your choice. Chetty then falls back as described above and keeps the rest of your settings; pick a new default when you notice.