Meeting recording policy

Meeting capture in KB2B happens from KB2B Desktop, an app that runs on your own computer (macOS today; Windows coming soon). We don't put any bot in the room — no extra participant shows up in Zoom, Google Meet or Microsoft Teams. Audio is captured locally on your Mac while you're in the call.

When the meeting ends, the app sends the audio to our recording provider (Recall.ai, EU) for transcription; our backend then extracts facts from the transcript and ingests it into your workspace's Knowledge POT. Citations in the chat link back to the exact moment of the meeting.

Our primary infrastructure and data storage and processing run in the European Union: the application on Fly.io in Frankfurt (fra) and the data on AWS eu-central-1 (databases, files, queues), with the vector index on Pinecone eu-central-1. Meeting capture and transcription run on Recall.ai in the European Union (Frankfurt, eu-central-1), and product analytics on PostHog's EU cloud. LLM processing (OpenAI, Anthropic) happens in the United States under their commercial API terms. Transitional note: frozen US backups from the migration are scheduled for deletion within 14 days of the cutover (August 2, 2026).

• Meeting audio: the recording provider's copy (Recall.ai, EU) is deleted automatically after 168 hours (7 days) — and immediately when you delete the meeting. KB2B keeps an archived copy in its own storage for as long as the meeting exists; deleting the meeting deletes it too.
• Raw transcript: stored within KB2B for as long as the meeting exists; deleting the meeting deletes it too.
• Facts extracted to the POT: kept for as long as the POT exists — they're the knowledge base you built, not the recording. They are deleted when you delete the meeting.
• Meeting metadata (title, date, attendees, duration): kept per workspace, deletable from the Meetings tab.

Only members of the workspace can access transcripts and extracted facts. We don't share data with third parties beyond the sub-processors that operate the service: the infrastructure described above and the AI providers we use for transcription and fact extraction. Under their commercial API terms, these providers do not train their models on your data; they may retain API inputs for a limited period for abuse monitoring.

It's your responsibility — just as it would be with a conventional recorder or any other note-taker (Granola, Otter, Fathom) — to inform meeting participants that the conversation is being recorded and transcribed.

Several jurisdictions require explicit two-party consent (California CIPA, various US states, GDPR in the EU depending on context). Before public launch we'll ship notice templates per jurisdiction; in the meantime we recommend a clear announcement at the start of the meeting:

“This meeting is being recorded and transcribed for internal knowledge management. If anyone prefers I pause the recording, let me know.”

• Stop the recording: from KB2B Desktop you can stop capture at any time. The audio captured up to that point is still processed (transcribed and turned into facts) — if you do not want it kept, delete the meeting afterwards (next bullet).
• Delete the meeting afterwards: from the Meetings tab you can delete a processed meeting whenever you want. This permanently deletes the transcript and all extracted facts (including their search embeddings) from your knowledge base, the archived audio and raw transcript from our storage, and the provider's copy on Recall.ai. Deleted data also leaves our backups as they expire, within 35 days at most.

Questions or complaints: contact@scipot.ai.