Usage terms

Rights to the audio

You may open a session only for audio you have the right to capture and transcribe. Where the law requires notice or consent from the people in the conversation, you provide it before capture starts.

Do not put secrets in participant names, agent names, or the audio-source label. Agent names become filenames. The audio-source label is stored with the summary. An RTMP URL is not logged, but it is still a credential if the URL itself grants access.

How a session is allowed to run

Sessions are realtime only. Capture starts after the speech socket is open. Audio is mono 16 kHz AAC, buffered for ten seconds by default, then sent. The same bytes are saved to the recording.

Only finalized turns are appended to the transcript. SRT cue times step one second per finalized turn. They are not the clock time of the recording. Speaker labels are the participant names you pass. They are not a guarantee of who spoke.

A session stops on the end key, on the configured duration, or on a stop-request file. The default duration is 60 minutes. Passing a duration in seconds wins over the minute setting. Ending the session flushes remaining audio, sends Terminate to the speech connection, and then writes one file per agent.

Keys and outputs

A live session requires your speech key. Agent files require your OpenAI key at the end of the session. If that key is missing, the recording and transcript may already exist and the agent files will not.

notes.txt and summary.txt follow the product's fixed sections. Any other agent tag is written as plain text. session.json stores paths to those files, not the transcript body.

Questions about Resum go to williams@viroke.com. Viroke Technologies Inc replies to enterprise inquiries from that address.