Video search for robotics and remote operations teams

Find the moment the run went wrong.

Vivu indexes the operational video your fleet already records, so a reviewer can pull candidate moments from an incident by describing it — instead of having to know the unit, the run and the timestamp first.

Start on the free plan. Demos run live on your own recordings.

"unit stopped mid-crossing"3 CANDIDATE MOMENTSrun_4471 · unit-12VISUAL: crosswalk · kerb · haltON-SCREEN: "STOP ENGAGED"00:12:38run_4318 · unit-07VISUAL: kerb · handoff · stall00:03:51run_3980 · unit-12VISUAL: crosswalk · halt00:47:06
01 / THE PROBLEM

You know it happened. You don't know where it is.

Telemetry tells you that something happened. Only the video tells you what happened. But before anyone can watch it, they have to work out which unit, which run, and roughly when — and then scrub a timeline until the seconds show up.

As fleet hours grow, the cost of every reconstruction grows with them. Reviews get shorter than they should be, or they get skipped, or the same event is worked through twice by two people who never learn of each other. And when the operator who knew which recordings mattered moves on, the judgement built up over hundreds of reviews leaves with them.

INCIDENT REVIEWTELEMETRY · SOMETHING HAPPENEDRECORDINGS · HOURS PER UNITunit-07unit-12unit-19?SOMEWHERE IN HOURS OF RECORDING.
02 / THE FIX

Search your recorded operations in plain language

Vivu is an AI search layer for the operational video you already record. Upload the recordings to a Vivu project and it indexes them there, without anyone reorganising the archive first.

It reads the objects and actions in the frame, what was said, and the text shown on screen. Ask for an event the way you would describe it to a colleague, and it returns candidate moments, each with its timecode and the recording it came from.

Fleet recorderOnboard camerasRun archiveALREADY RECORDED. UPLOADED AS-IS.INDEXVISUAL · SPEECHON-SCREEN TEXTANYONE REVIEWING"unit stopped mid-crossing"CANDIDATE MOMENTS, WITH TIMECODES

Evidence instead of scrubbing

Describe what happened and get candidate moments back, each carrying the recording it came from and its timecode. The review starts from the event rather than from a file name.

The review outlives the reviewer

The index belongs to the team. When the person who knew which recordings mattered moves on, the ability to find things in the archive stays behind.

One incident becomes a pattern

The same kind of moment can be pulled across separate incidents, so a single reconstruction turns into material for QA and for training remote operators.

It stays out of the safety path

Vivu reads recordings that are already stored. It is not in the live control loop, it makes no driving or safety judgement, and the conclusion is drawn by your reviewers.

03 / WHERE IT FITS

Where Vivu fits in an incident review

When an incident is flagged, the review starts from what happened rather than from a file name. Describe the event and candidate moments come back. Watch the seconds that matter, with the source recording and the timecode beside them. Then the same evidence feeds QA and training, instead of being watched once and closed.

ML and data teams working on robots or autonomous vehicles can use the same search on recorded runs: describe a grasp that slips or an unprotected left turn in the rain, get candidate clips back, and have a person check each one before it goes into a training, evaluation or QA set.

  1. Incident triggered

    Start from what happened

    Not from a unit number, a run ID and a guess at the time.

  2. Narrowing the window

    Describe the event, get candidates

    Plain language in; candidate moments out, each with its timecode.

  3. Reconstructing it

    Watch the seconds that matter

    The source recording and the timecode travel with every moment.

  4. Closing the loop

    The evidence gets reused

    The same moments feed QA and training instead of being watched once.

04 / FAQ

Does Vivu touch our real-time or teleoperation stack?

No. Vivu reads operational video after it has been recorded and stored. It sits outside the live control path, and nothing in a review session reaches a robot.

Does Vivu decide what went wrong?

No. Vivu returns candidate moments with their timecodes and source recording. Reading them and deciding what happened stays with your operations and safety reviewers.

Do we have to re-encode or reorganise our recordings?

No. Recordings in any common video format go into a Vivu project as they are, with no reorganising step, and every result points back at the recording it came from and its timecode. Camera streams kept inside log formats such as ROS bag or MCAP need to be exported to video files first.

Start free, or bring us the recordings.

Upload a few recordings on the free plan, or talk to us about the operational archive.

A demo is 30 minutes, on your footage. Contact Sales takes three fields and hands you the calendar.

Upload only video you are authorized to share and process. See our Terms and Privacy Policy.