Keep the video. Find the moment that matters.

Save the original with the work

Add the video through the Session attachment button or Project Files. Large uploads use parts: you can pause and resume, and select the same file again after an interruption to continue within 24 hours. Workspace checks the complete file before reusing saved parts.

The original remains a Workspace file after processing. Google’s temporary processing copy is separate from your saved original. Uploading a file does not by itself request a paid video analysis.

Choose the limits that fit your recordings

Limits below use decimal MB, GB, and TB. The storage pool belongs to the Project owner and is shared across their Workspace. Originals, retained file revisions, and processing copies consume storage. Uploads and work in progress reserve space before they begin.

An account above its allowance keeps access to existing files. New uploads need room within the current plan. Deleting a visible file may not immediately remove retained revisions.

Set a question, interval, and credit limit

Choose transcript only, summary, detailed analysis, or a question about the video. A request can cover up to four hours; select an interval for longer recordings. Work continues in the background, and the preview shows its current status. Preparation reserves credits before it starts; its recorded processing time and completed model calls use credits. Unused reserved credits are released. Cancelling stops further work.

The Session’s main model can request the same video tools. Processing uses Gemini 3.8 Flash for visual understanding and Gemini 3.5 Transcribe for speech. Accounts with a protected inference boundary cannot silently send videos to the shared provider. Availability is shown before starting.

Review words, visuals, and interpretation separately

A spoken claim that a deployment succeeded is different from a screen showing a successful deployment. The transcript records spoken words; observed visuals describe the picture; interpretation cites the supporting evidence. Click a timestamp to inspect the original moment.

Audio is split into bounded sections for transcription. Speaker labels distinguish voices within each section and do not establish a person’s identity across sections. Check important quotations, names, and claims against the original. Explicit coverage tells you which intervals were completed; partial processing cannot establish what happened elsewhere.

Bring channel videos into the same Workspace

Supported WhatsApp and Viber video messages are imported into their connected Session, with captions preserved. Connected Telegram bots use the same saved-file flow and retain their allowed-sender rules. Group mention rules still decide when an Agent replies.

Channel platforms can impose smaller transfer limits than your Workspace plan. Hosted Telegram bots can download up to 20 MB. If a platform cannot provide a video, open the private Workspace upload link in the Session and upload the original there. The WhatsApp linked-device bridge remains unavailable.

Reuse the result with timestamps intact

Search the transcript or observed visuals in the video preview. Saved transcript and visual evidence also appear in Project text search with timestamps. Export the result as JSON, plain text, Markdown, SRT, or VTT for review and reuse.

Keep the source revision with the result when sharing it. A later edit or replacement of the video needs a new analysis. A structured JSON result validates the shape of the output; it does not prove that every observation or transcription is correct.

Try it with a small example.

Before you use the result.

• The original opens from Project Files.
• The reported coverage matches the intended interval.
• Quoted speech has been checked against the video.
• Interpretation cites visible or spoken evidence.
• Exports keep the original timestamps.