Models and workload routing
Settings → Models → Workload routing decides which connection, model and reasoning effort each kind of work uses. Changes apply to new turns.
| Workload | Key | Used for | If unset |
|---|---|---|---|
| Conversation | chat | Talking with you, on every channel | Required |
| Triage | triage | Mail, messages and events in the background | Conversation's model |
| Follow-ups | wakeup | Scheduled wakeups, routines, the morning plan | Conversation's model |
| Sensitive conversations | sensitive | Sensitive mode, or adult content found by Jev | Conversation's model |
| Simple requests | simple | Requests Jev grades simple | Conversation's model |
| Complex requests | complex | Requests Jev grades complex | Conversation's model |
Reasoning: Provider default, Off, Minimal, Low, Medium, High, Extra high or Max, where the model supports it. Higher effort is slower and dearer; use it for Complex requests rather than for everything.
A connection marked not connected in the list must be connected first (Choose a model).
Routing with Jev
With a TypeSafe key connected, Jev can read the last few messages before each of your turns and pick the model:
- Adult content to the sensitive model — uses the Sensitive model and instructions when the conversation is adult content.
- Route by complexity — simple requests to the Simple model, complex ones to the Complex model, the rest to Conversation.
Both are off by default. If Jev does not answer, the usual model runs. Your messages go to TypeSafe for this.
Keeping costs down
- A small fast model for Triage: it runs on every batch of new mail and messages.
- Large tool families load only when needed (What your accomplice can do), which keeps each request small.
- Goals and initiatives have a daily token budget (400,000 by default) on the Initiatives page.
- Initiatives shows background tokens today and your conversation's tokens over 24 hours.
Sign-ins
ChatGPT sign-ins (the OpenAI connection on your ChatGPT plan) and xAI sign-ins renew automatically while their refresh token is valid; Settings → Overview flags an expired one. xAI can be signed into again from its page (Sign in again). A new ChatGPT sign-in cannot yet be started from the Workspace; an OpenAI API key replaces it if it ever expires.