3 April 2026
Bot Rejoin, Fact-Checking, Analytics & AI Self-Reflection
The biggest release yet. Bots can now rejoin interrupted meetings and merge sessions into a single transcript. Fact-checking with inline source citations lands in the minutes editor. A new Analytics Dashboard gives admins full visibility over team usage. Plus: quality gates, self-reflection loops, and dynamic model selection.
NEW FEATURES
Multi-Session Meeting Support & Bot Rejoin
Why we updated this
If a meeting bot dropped out mid-call or a meeting resumed after a break, there was no way to bring the bot back in. The original session was lost and users had to start from scratch.
How it works now
You can rejoin the bot to any meeting directly from the calendar or the meeting sidebar. Each rejoin creates a new session, and all sessions are merged into a single transcript with clear session separators showing timestamps. The 'Rejoin Bot' button appears automatically when a bot has disconnected.
What to do
No action needed — the Rejoin Bot option appears automatically in your calendar event actions and meeting sidebar.
Fact-Checking & Source Citations
Why we updated this
When editing minutes inline, there was no way to verify whether a statement was supported by the transcript or knowledge base. Edits were made on trust.
How it works now
The AI now fact-checks inline edits against your transcript and knowledge base documents. Source citations appear as inline badges — hover over them to see the document name, page number, and relevant extract. Unverified claims are flagged so you can review them before finalising.
What to do
No action needed — fact-checking runs automatically when you make inline edits.
Analytics Dashboard
Why we updated this
There was no centralised view of meeting activity, minutes output, or team usage. Admins had to piece together usage data manually.
How it works now
A new Analytics section shows KPI cards with trend indicators, sparklines, completion gauges, and meeting type breakdowns. You can filter by team, meeting type, and date range, and export filtered results as CSV. Organisation owners see all teams; team admins see only their own.
What to do
Navigate to the new Analytics section in the sidebar to explore your team's data.
Custom Vocabulary Extraction
Why we updated this
Industry-specific terminology, acronyms, and proper nouns were often misrecognised in transcriptions. There was no way to teach the system your organisation's vocabulary.
How it works now
In Meeting Setup, you can now extract custom vocabulary from your selected knowledge bases. The system identifies specialised terms and injects them into the transcription pipeline, improving accuracy for your organisation's specific language.
What to do
Optional — go to Meeting Setup, select a knowledge base, and click 'Extract Vocabulary' to improve transcription accuracy.
IMPROVEMENTS
Minutes Quality Gates
Why we updated this
Generated minutes sometimes had inconsistencies: decisions mentioned but not resolved, glossary terms used incorrectly, or minor style issues like missing em-dashes.
How it works now
The minutes generation pipeline now runs three deterministic quality gates — checking decision resolution, glossary term usage, and style compliance — before delivering the final output. These checks run without additional processing cost and catch issues that the AI reflection loop might miss.
What to do
No action needed — quality gates run automatically during minutes generation.
Locked Text Preservation
Why we updated this
Boilerplate sections in minutes templates — such as standard disclaimers or standing headers — were sometimes modified during generation, requiring manual correction.
How it works now
Sections you mark as locked in your template are now fully preserved. The minutes generation pipeline respects these boundaries and will not alter locked content under any circumstances.
What to do
No action needed — existing locked sections are automatically respected.
AI Self-Reflection Loop
Why we updated this
First-pass minutes generation occasionally missed nuances or produced inconsistent formatting across sections.
How it works now
The AI now performs iterative self-reflection, reviewing and refining generated minutes through multiple passes before delivering the final version. This produces more polished, consistent output without requiring manual editing.
What to do
No action needed — self-reflection is enabled automatically.
Shared Device Detection & Prompt
Why we updated this
When multiple participants joined from the same device (e.g., a conference room), the system could not distinguish individual speakers, reducing transcription accuracy.
How it works now
A shared device prompt now appears in the meeting sidebar, allowing you to confirm or revoke shared device participation. Physical speaker names are tracked to improve speaker attribution during transcription.
What to do
Optional — when prompted, confirm shared device participation to improve speaker identification.
Dynamic Model Selection
Why we updated this
Available AI models were hardcoded, meaning new models required a platform update to become selectable.
How it works now
Models are now fetched dynamically based on your organisation's configured API keys, with support for the latest model variants. If the API is unavailable, the system falls back to a static list.
What to do
No action needed — new models appear automatically based on your API key configuration.
FIXES
Transcript Rendering & Timestamps
Why we updated this
Transcripts with invalid timestamps or non-standard formats occasionally failed to render, showing blank or broken content.
How it works now
The transcript viewer now gracefully handles invalid timestamps and supports both structured (JSON) and plain-text speaker formats. Skeleton loading states appear while transcription data is loading, replacing the previous blank screen.
What to do
No action needed.
British English Date & Time Formatting
Why we updated this
Dates and times across the platform were inconsistently formatted, sometimes appearing in US format (MM/DD/YYYY) rather than the expected UK format.
How it works now
All dates and times now default to en-GB formatting throughout the application, matching the platform's British English standard.
What to do
No action needed.