Source formats and limits
The preview accepts UTF-8 text, not recordings or document binaries. Files are decoded and parsed in a local worker. Their content is not sent to a model or uploaded by this app.
Supported formats
- Plain text / article: paragraph-based source text. Paste or import a TXT file.
- Speaker transcript: speaker-labelled TXT, with optional timestamps. A start timestamp alone is insufficient for timed-caption export.
- SRT: numbered cues with start and end times.
- WebVTT: a WEBVTT header followed by timed cues; supported voice tags are retained as speaker metadata.
- JSON: a strict object with a segments array. Each segment has text, speaker, start, and end. JSON times are seconds with up to three decimal places. Use null for unknown timing.
Choose the matching format when importing. The file takes precedence over pasted text. Source names help you identify excerpts and appear in private evidence exports.
Structured JSON example
{"segments":[{"speaker":"Maya","start":1,"end":5.5,"text":"Good work starts with a clear conversation."}]}Practical limits
One workspace supports 8 collections. Each collection supports 8 sources, 40 assets, and 10 saved releases. A source is limited to 1 MB of UTF-8 text, 250,000 characters, and 2,000 segments (up to 8,000 characters per segment). An asset can hold up to 100 blocks. The restorable workspace backup is limited to 12 MB, including source copies inside releases. Device storage may run out before that limit.
Sources retain their original imported text and a SHA-256 fingerprint of that text. This fingerprint detects changed text; it does not prove who recorded it or authenticate the original file bytes.
What is not supported
PDF, DOCX, audio, video, remote URLs, automatic transcription, and live channel imports are not accepted by this preview. Convert or transcribe your material with an appropriate tool, check its accuracy, and import the resulting text. Do not paste secrets or material you do not have permission to process.