Calendar lookup in the assistant, new Library tabs, and sturdier uploads and deployments
· 2 min read
The assistant can look up your calendar, the Library gains Videos and Images tabs, and uploads, voice, and deployments are more reliable.
The assistant can now answer questions about your calendar, and the Library has dedicated Videos and Images tabs. File uploads, voice conversations, and deployments are also more reliable, and several security checks have been tightened.
Calendar in the assistant
- Ask the assistant for your meetings today, this week, or on a specific date. Agendas appear as cards right in the chat, with timezone support.
- System Agents has a new Calendar section for choosing which agents respond when Google Calendar events are created, updated, or deleted. Workspaces without one configured are skipped cleanly, and custom hook overrides are now recognized correctly.
Library and media
- The Library now has Videos and Images tabs. Creatives focuses on generating content, and generated images can be searched and filtered.
- Generated images get stronger privacy controls: data export, erasure, and automatic cleanup when a workspace is deleted.
Voice
- The voice agent supports newer ElevenLabs voice models (v3 and newer).
- Chat voice settings are now in the composer popover at the bottom of the chat, with the full speed range restored. Voice turns recover from empty or failed responses by retrying automatically, and the assistant routes work more accurately when it confirms that coding work has started.
Uploads
- Uploading rubrics, data rooms, and audio samples no longer fails with "Failed to fetch". When an upload does hit a connection problem, you get clear guidance instead of a generic error, and Library uploads only accept supported file formats.
Releases, security, and stability
- Major updates now require an approved, completed Implement pass before a pull request can merge or deploy.
- Release ramp steps and the bake phase can be given low-traffic extensions, so quiet periods don't cut releases short.
- Deployments that hit rate limits are parked and retried automatically instead of failing. GitHub capacity limits on pull requests now show up as retryable responses rather than server errors.
- Stricter validation of Jira and Confluence identifiers closes request-forgery risks.
- Agent starts no longer get stuck in a pending state. Terminal reservations recover from dropped connections, pull request creation stops retrying when a branch is missing, and sessions on a signed-out Muse CLI are detected and re-routed.
Plus 12 smaller improvements, and 21 earlier changes not covered here.