Choose models and set roles
Decide which model answers by default, which one keeps the housekeeping going, and what each account may pick.
Choose and switch models
A model is identified by its provider and model ID, written <provider>/<model>. A saved choice is restored while that pair stays configured and allowed.
If the pair becomes unavailable, Chetty uses the applicable Project, account or server default. It does not silently match the same model ID under another provider.
Choosing a model does not change your Project, tools, plugins, memory identity or permissions.
Set the instance roles
Settings → Models → Model roles answers which model does what for the whole instance.
| Role | What it does |
|---|---|
| Default chat model | What a conversation starts with when the account has no personal choice. It is stored independently of provider and catalog order, and account permissions still apply. |
| Utility model | Writes conversation titles and processes long-term memories. |
| Daily digest model | Writes the dashboard summary. It inherits the utility model unless you set one. |
If the stored default becomes unavailable, the row keeps it visibly unavailable and conversations use the existing permitted-model fallback. Reconnecting the provider does not overwrite your choice.
Manage the catalog
Settings → Models holds the catalog: which models exist, context-window overrides and per-model output-token limits. Refresh catalog reloads it from the providers.
- An auto-discovered catalog keeps only models reported with text output, and drops OpenRouter's zero-cost
:freevariants. - A model an administrator lists manually is kept regardless of its reported kind.
Context size comes from the provider's own information or from an administrator override. Input and output share the window; Chetty leaves room for the prompt and uses 8,192 output tokens by default when no limit is supplied.
Personal model settings
Every account sets its own chat, image and context-shortening models in Account → Models, and its default there decides what new conversations start with. Administrators can use the instance catalog; other accounts can choose only models the administrator allows and their own policy permits.
- In web chat, use the model picker in the status line.
- In
elowen chat,/modelopens the picker and/model <model>switches directly. Ctrl+P inside the picker opens provider management. - Account → Models also sets a per-model compaction threshold beside the global one.
Tip: The same picker sets reasoning depth and fast mode. Read Model and reasoning depth.