Release Notes: 2026-08-31

Accurate engineering metrics, a Get Documents tool for connected assistants, and self-healing meeting recordings

What's new

Accurate Cursor and company-wide metrics

• Cursor usage data (accepted code suggestions, tab completions, and agent requests) now populates reliably across the AI-Assisted Engineering card and Metrics Explorer, instead of showing blank or stale numbers.
• Company-wide totals, like total licensed seats, now show today's actual count in Metrics Explorer instead of an inflated sum across the selected date range.
• Team-wide adoption rates, like Cursor's suggestion-acceptance rate, are now calculated correctly instead of overstating actual usage.
• Asking Bloomy about company-wide numbers, like total Cursor seats or GitHub activity, now returns an answer instead of "no data found" for every company.
• Task and reminder mentions in Bloomy notification emails now render as clickable links to the task, instead of raw text markup.

DORA counts merge-based deployments correctly

• Repos where deployment happens automatically when a pull request merges, rather than through a separate release step, can now be set to count merges as deployments, so deployment frequency is no longer undercounted.
• Deployment rows in the DORA view now link directly to GitHub, with the pull request title shown inline.
• A crash that could interrupt DORA reporting for companies with data split across two regions is fixed.
• Bloomy's DORA answers in chat now reflect the same per-repo deployment setting shown on the dashboard.

Full document retrieval for connected assistants

• MCP API keys can now grant access to a Get Documents tool, so a connected assistant can pull the full content of a specific transcript, coding session, or upload after finding it in search.
• Filtering search results by document type now actually filters, instead of silently ignoring the filter and returning mixed results.
• Paging through a long list of documents no longer risks looping back to the start if a page token gets corrupted.
• External MCP clients can now request up to 200 results per page instead of 50.

Meeting recordings recover automatically

• Google Meet transcripts that previously showed raw system identifiers instead of participant names now show clear labels, and can pull in real names and meeting subjects from your calendar when available.
• A meeting recording from a brand-new or not-yet-matched device now recovers automatically within a day, instead of staying stuck until someone manually reruns it.
• The recommended macOS device agent (now 1.13.3) no longer lets itself update in the middle of an active meeting recording.

Who benefits

• Engineering leaders reviewing Cursor and DORA data: accurate usage, seat, and deployment numbers across Metrics Explorer, the dashboard, and Bloomy chat.
• Teams whose deployment process ships on merge: DORA deployment frequency is no longer undercounted, once configured per repo.
• Developers and admins connecting AI assistants like Claude Desktop over MCP: access to full documents by ID, and pagination that no longer loops on a bad page token.
• Anyone reviewing meeting recordings or transcripts: fewer stuck recordings, and participant names that show correctly on Google Meet calls instead of raw system identifiers.

Try it

• Review the AI-Assisted Engineering card or Metrics Explorer to see corrected Cursor and company-wide totals.
• Set a per-repo deployment signal in your DORA settings if your team deploys on merge.
• Grant the Get Documents tool to an MCP API key from your connector settings.
• Check Device Agents for the recommended macOS agent version (1.13.3).