This release added more economic and visual clarity. Model selection became easier to reason about before you commit, activity streams got better at understanding different coding CLIs, and notifications stopped flattening everything into plain text.
Model Switching Became More Informed
Inner-loop settings now give you more context before you switch an agent to a different model.
- Projected daily and monthly cost estimates now appear alongside model choices
- Relative change versus the current setup is surfaced directly, so you can compare options instead of guessing
- The current saved setting stays visible with its own estimate, which makes changes feel more deliberate
Activity Streams Understand More CLIs
Run logs are no longer so narrowly shaped around a single coding agent.
- Structured parsing now works across Claude, Codex, Gemini, and OpenCode
- Mixed CLI output renders more cleanly in activity views
- Inner-loop loading and status handling became steadier, which removes some of the friction around experimenting with different execution tools
Notifications Preserve More Meaning
Alerts became easier to skim because they can now keep more of the original structure.
- Email notifications render markdown instead of flattening it away
- In-app notifications can show headings, lists, and code more cleanly
- Heartbeat cards now refresh themselves more readily, so the current state is easier to trust without a manual reload