Public reader · Manage agendas Channel assignments · Direct HLS

Meeting capture

Operator tools. Capture starts privately; reviewed materials and speaker labels have explicit sharing controls. Meeting publication remains separate.

Sign-in stays active while this operator tab is open and connected. Capture traffic renews it too. After 24 hours without authenticated activity, sign in again. Browser session restore can preserve cookies; use Sign out to end access.

Meeting title Meeting/source ID (optional; same ID prevents duplicate capture) Capture limit (minutes)

Optional audiovisual activity analysis

Recorded / direct HLS analysis controls Analyze consecutive ten-second shared-tab A/V windows (private model worker required) Anonymous regions JSON (optional; seat/view IDs and normalized boxes, no names) I checked this source's A/V-to-transcript clock. Do not pause or seek while capturing.

Disabled. This does not automatically publish or identify a person. No queue; busy windows are skipped.

Save selected screen-text candidates (local OCR; review required) Automatically share extracted screen documents when this meeting is published (unreviewed; AI requires approval) Source playback offset at start (seconds)

Invented test audio only (60-second maximum)

Open the publisher player in another Chrome tab. Choose Chrome Tab and turn on Share tab audio. Microphone and whole-screen audio are not used. Keep this tab open; silence, interruptions and chunk boundaries need review.

Open the invented test clip

Audio activity

Ready for an audio check. Checking does not send audio to a provider.

Transcription sends short audio chunks to OpenAI and saves provisional text privately. Audio is not archived. Arrivals lag behind playback; timestamps mark chunk bounds, not individual words. No speaker identification is claimed.

Screen text — private review

Speaker evidence — experimental, private

Buffer silent video frames for automatic speaker evidence (requires screen sampling)

With the diarization transcription model configured, anonymous turn events automatically collect matching video frames and bounded OCR candidates. Other models require manual marking. Names and cross-batch identity remain unverified. Timing is approximate. Camera views and nameplates require review.

Anonymous speaker ID

Reviewed speaker labels

Label a finite saved passage from an explicit introduction or reviewed on-screen context. This does not identify faces or voices, or carry a name into later unreviewed turns. Boundary-straddling passages remain unchanged.

Session ID Start (capture seconds) End (capture seconds) Speaker name Evidence Show this reviewed label in the published transcript

By using this agent, you agree to our Agentic Policy.