Intelligence in the Chat
Analyze videos with your lenses, ask them questions with cited answers, and design, publish, and activate lenses by talking to your AI assistant.
Intelligence in the Chat
Everything you do with VIDTREO Intelligence in the dashboard, you can also do from your assistant: run a lens on a video, read the result, ask the video questions, and create or change the lenses themselves. Analyses started from the assistant are the same analyses — they appear in the dashboard, are charged the same, and fire the intelligence.lens_analysis.completed webhook.
Through the MCP, analyses work on videos. Analyze audio recordings from the dashboard; their transcripts are available to the assistant with get_audio_transcript.
Analyze a Video
"Analyze the Acme discovery call with the Sales discovery lens"
The assistant finds the video (search_videos) and the lens (list_lenses), and asks you to confirm when it is unsure which one you mean.
It calls start_analysis. The analysis is charged per analyzed minute — see Pricing — so the assistant only starts one when you asked for it.
The analysis runs in the background. The assistant checks it with list_video_analyses until its status is completed.
It reads the result with get_analysis_result — summary, sections, data points (findings), chapters, and scorecard — and answers you from it.
Analyses started from the assistant behave like manual analyses with the default inputs: Spanish and English as the spoken language, with no question and no participants. The video must have finished processing, and the lens needs a published version. Which version runs is explained in Which version analyzes a recording.
The result has the structure described in the Result reference, without the word-by-word timings — the assistant reads the transcript with get_video_transcript when it needs it.
Ask a Video
"In the Acme call, did they mention a budget? When?"
ask_video answers a question about one video from its latest completed analysis, with citations to the moments that support the answer — the same as Ask on the recording page of the dashboard. To answer from an older analysis, the assistant passes its id.
- The video needs a completed analysis. If it has none, the tool says so and the assistant offers to analyze it first.
- Asking about a single recording needs a role that can create videos (Owner, Admin, or Developer) — see Permissions.
For questions that only need the words spoken, the assistant can also read the transcript directly, with no analysis needed.
Create a Lens
"Help me create a lens to screen candidates for the SDR role"
The assistant designs the lens with you, the way the dashboard's Designer does. It never creates a lens from a single sentence:
Interview. It asks a few questions per round, with options it suggests from what it knows about your company — the conversation, its memory or project context, your existing lenses (so it does not duplicate them), and your recordings when useful. It covers: which recordings the lens is for and what decision it supports, what to extract, whether and how to score, the summary sections, the output language, what the analysis must never assume, and — for videos — what to check on screen. It follows up while an answer is vague.
Draft. It writes the whole lens — name, objective, summary sections, data points, rubric, trust rules, visual checks — from your answers.
Review. It shows you the complete proposal. You approve it or ask for changes, as many times as needed.
Decide. It asks whether to publish the lens now or keep it as a draft to review in the dashboard, and — when publishing — in which environments to activate it. It does not assume either: if you approve without choosing, it asks again.
The lens is created with every building block described in Lenses, within the platform limits:
| Building block | Configuration key | Limit |
|---|---|---|
| Objective | objective | 4,000 characters |
| Instructions | instructions | 8,000 characters |
| Output language | outputLanguage | es, en, or auto |
| Summary sections | summarySections | 20 |
| Data points | fields | 50 |
| Rubric criteria | rubric (+ scale) | 30 |
| Trust rules | trustRules | 30 |
| Visual checks | visualChecks | 20 |
A draft is never activated and is not used by analyses until it is published. Publishing validates the configuration with the same rules as the dashboard.
Edit a Lens
"Add a data point for the candidate's notice period to the SDR lens"
- The assistant reads the lens with
get_lens, including every version and its configuration. - It changes only what you asked for and shows you the new configuration.
- It asks whether to publish the change now or leave it as a draft, then calls
update_lens_config, which saves a new version — earlier versions are always kept.
Publishing — from update_lens_config or later with publish_lens_version — also moves every environment where the lens is active to the new version, so the next analyses use it. Analyses already queued or running keep the version they started with.
| If this fails | The assistant tells you |
|---|---|
| Publishing the new version | The version was saved as a draft, and analyses still use the previous one |
| Moving the activations to it | The version was published, but the environments still use the previous one — publish_lens_version retries it |
Renaming a lens (rename_lens) does not create a version.
Where a Lens Runs
| Ask | Tool |
|---|---|
| "Which lenses run in the Sales environment?" | list_lens_activations |
| "Activate the SDR lens in Interviews" | activate_lens |
| "Stop using it in Interviews" | deactivate_lens — the lens and its analyses are kept |
| "Analyze every new interview automatically" | set_lens_auto_analyze |
Activating a lens from the assistant leaves auto-analyze off, unlike the dashboard. Turning it on analyzes every new recording in that environment and charges each analysis, so the assistant only turns it on when you ask, after telling you about the cost. Turning it off is always safe.
Delete a Lens
"Delete the old onboarding lens" — the assistant asks you to confirm, naming the lens, before calling delete_lens. The lens is archived: it disappears from the list, can no longer be used or activated, and cannot be restored from the assistant.
Compared to the Dashboard
| Dashboard | Assistant | |
|---|---|---|
| Design a lens | Designer, from a sample recording or a description | Interview in the chat, using what the assistant knows about your company |
| Auto-analyze when activating | On by default | Off until you turn it on |
| Analysis inputs | Spoken language, question, and participants | Defaults (Spanish and English, no question, no participants) |
| Recordings that can be analyzed | Videos and audios | Videos |
| Ask | One recording, or across recordings (Ask your recordings) | One video at a time |
| Compare versions, roll back | History tab | Reads every version with get_lens; to go back, it saves an earlier configuration as a new version |