Sessions
See where context goes and keep every thread navigable.
#Read the breakdown
- Input: tokens sent to the model, including system prompt and history.
- Output: tokens the model generated in this session.
- Reasoning: thinking tokens spent before the visible answer.
- Cache: cache-read and cache-write tokens that lower repeat cost.
- Summary: tokens held by the session summary after summarization.
The footer always shows context percentage and cumulative cost, so stop guessing and check the meter.
#Summarize long sessions
Long sessions are summarized, not compacted. The summary becomes a distinct segment in the breakdown, and token estimates stay accurate after summarization.
#Thread the work
Organize work as parent sessions with child threads. Delegate parallel explorations to children while the parent keeps one navigable history.
#Manage sessions
#Filter and bulk-act
Filter the session manager to single-row results, select many sessions, and run bulk operations with full keyboard navigation.
#Delete optimistically
Delete a session and the card disappears immediately. If the delete fails, the manager surfaces recovery actions instead of dead-ending.
#Sync across clients
Switch the active model, agent, or effort variant on desktop, TUI, or CLI and it follows you everywhere. A storage watcher streams session and message updates across processes in real time.