Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.kb2b.app/llms.txt

Use this file to discover all available pages before exploring further.

This page covers the web side of meetings — what you see in kb2b.app/dashboard/meetings/[id] after a recording finishes. To record meetings from the desktop, see kb2b Desktop.

​
The full cycle

Recording ends           →  status: recording
   ↓
Upload to backend       →  status: transcribing
   ↓
Transcript ready       →  status: done
   ↓                       ↓                     ↓
   ↓                Claude summary         Facts via SciPot
   ↓                (per-language)         (async ingestion)
   ↓                       ↓                     ↓
   ↓                Highlights extracted   Edges + scores
   ↓                                       Contradictions detected
When you open /dashboard/meetings/[id], what you see depends on status. While transcribing, you see a placeholder; once done loads summary, facts, and chat in parallel.

​
Meeting summary

kb2b generates a full markdown summary of the meeting using Claude. Features:
  • Per-language. If your team works in Spanish but a customer watched the meeting in English, you can generate the summary in both. kb2b caches each language separately (summariesJson.es, summariesJson.en).
  • Structured. Includes context, participants, agenda covered, decisions made, next actions, a ## Highlights section, and optionally notable sentiments or non-verbal aspects.
  • Cached. Once generated, it isn’t regenerated unless you switch language or request a manual refresh.

​
Highlights

Highlights are the meeting’s key points — auto-extracted by kb2b from the ## Highlights section of the markdown summary:
  • Typically 3–7 highlights per meeting.
  • Each one with a timestamp pointing to the exact moment in the transcript.
  • If the meeting has no manual title, kb2b uses the first highlight as the title in the meetings list.
  • Highlights are NOT standalone POT facts — they’re narrative summaries of the specific meeting.

​
Facts extracted to the POT

Apart from the narrative summary, kb2b ingests the transcript into the POT as another document, and SciPot extracts structured facts:
  • Each fact has its POT Score and level (Extracted, Inferred, etc.).
  • Each fact points to exact citations in the transcript (provenance: filename + offset).
  • If any fact contradicts a Constitution axiom, the contradictions workflow kicks off (see POT Constitution).
  • Facts inherit the tags and project assigned to the meeting.
On the meeting detail page, the “Extracted facts” section loads lazy (on scroll or expand). Gives you a quick view of what kb2b now knows that it didn’t before.

​
Chat about the specific meeting

Each meeting has its own chat, separate from the POT’s global chat:
Chat typeContextWhen to use
POT chat (/dashboard/chat)All facts in the accountBroad questions: “What’s the overall situation with Acme?”
Meeting chat (/dashboard/meetings/[id])Only facts extracted from that meetingSpecific questions: “What did we decide about pricing on this call?”
Meeting chat scopes context to that meeting’s facts so it doesn’t blend with prior knowledge. Useful for preparing follow-ups or clarifying what was decided.

​
Statuses you can see

StatusWhat it meansWhat to do
recordingRecording is running on the desktopWait
transcribingRecall.ai is processing the audioWait 2-5 min typically
doneReady: summary + facts + chat availableReview and interact
failedSomething failed in transcription or ingestionCheck the error on the page; report to support if it persists

​
If the summary doesn’t generate

If status is done but the summary takes more than 30 seconds:
  1. Refresh the page.
  2. If still empty, check /dashboard/consumption — your workspace may be at 110% tokens (hard limit blocks LLM operations including summaries).
  3. If everything looks OK, email support@kb2b.app with the meeting ID.

​
Retention policy

Transcripts are kept while your workspace is active. Raw audio is deleted after 7 days (Recall.ai free tier). Extracted facts in the POT are never deleted — they’re the account’s persistent memory. Legal details in Recording policy. Page in progress Phase 1 — screenshots of summary and facts panel coming in Phase 2.