This release tightens the loop between execution and user interaction: notifications are easier to manage, agents can introspect their own activity feed, and interactive steps are consolidated so you do not have to juggle multiple threads.
Notifications: Clear All + Better Context
Notifications should help you act, not accumulate as background noise.
- Clear all notifications - One click clears the currently visible list (with an optimistic UI that rolls back safely on failure)
- Read without losing history - Opening the popover marks notifications read (badge clears), while keeping the list visible for review
- Better "new chat" behavior - Triggers are unified and new chat responsiveness is improved
- Email notifications include agent name - Easier to understand what (and who) the notification is about at a glance
New Agent Tool: getActivityFeed (For Debugging And Status)
Agents can now query their recent activity feed as structured data. This helps them answer questions like "what failed?", "what changed?", or "what happened last?" without guessing.
What The Tool Can Return
- RUN items - execution attempts (started/succeeded/failed) with timing, inputs, and outputs
- VALIDATION_RECORD items - schema validation outcomes and severity
- REPLANNING_EVENT items - plan revisions with before/after snapshots
- CRUD_EVENT items - entity create/update/delete events with change context
Filters And Pagination
- Filter by activity type and/or entity ID
- Cursor-based pagination for long histories
- Large run fields are automatically trimmed to keep responses useful (not token-bloated)
updateEntity Tool: More Expressive, More Correct
Agents can now update planning entities more precisely and with fewer schema gotchas.
- More interactive control - Agents can update interactive step prompts, choices, links, and structured validation more precisely
- CHOICE/CTA guidance included - Clearer tool descriptions for interactive modes
- More consistent execution mode handling - Fewer mismatches between what is configured and what actually runs
Unified Interactive Threads (ASK_USER, CHOICE, CALL_TO_ACTION)
Interactive steps are now coordinated through a shared thread when they belong to the same orchestration.
- One thread per orchestration run - Interactive prompts share a single thread so you can reply in one place
- Idempotent thread creation - Concurrency-safe creation prevents duplicate threads
- Consistent message format - Interactive prompts post as assistant messages with deep links back to the relevant planning entity
- Unified notifications - In-app/email notifications link you directly to the shared thread
Result: fewer scattered "where do I reply?" moments and smoother execution when a plan needs multiple decisions.
Small But Important UX Polish
- Assistant message actions stay visible - Actions remain available even after interacting with a message, reducing UI friction