We use essential cookies for authentication and optional analytics cookies to improve the platform. Privacy Policy

    VidForgeX · video intelligence

    Your video already knows the answer.

    Companies record everything — security cameras, screens, meetings — and watch almost none of it. VidForgeX reads the footage so people don't have to, and every answer points to the exact recording and moment it came from.

    Ask your footageone engine · every kind of video

    Security camera

    Screen recording

    Meeting

    Why is claim

    Two products, one engine

    Built for the two places video piles up fastest.

    Security footage that nobody can afford to review, and recordings of work that nobody has time to study.

    ForensicFor investigators and security teams

    Nobody should have to watch the footage.

    Upload recordings from every camera and search them for a person, a vehicle, a plate or a voice — by photo or in plain words. Results are ranked, explained, and one click from the moment.

    • Find a person or vehicle across every camera at once
    • Licence plates, left-behind bags, watchlist alerts
    • Fingerprinted evidence and a permanent record of every decision
    Time & MotionFor operations, back office and engineering teams

    See how work really gets done.

    Turn screen and floor recordings into timed steps, cycle times and bottlenecks — and let Viddy, the AI meeting teammate, write the minutes and file the follow-ups.

    • A time and motion study without a stopwatch or an observer
    • SOP checks and SOP drafts from how the work is really done
    • Viddy joins Zoom, Meet and Teams; tasks land in Jira, Trello or GitHub

    Under the hood

    Same engine, whatever you point it at.

    Every product runs on one pipeline, so an improvement to how we read footage reaches every customer at once.

    Every frame, screened for motion
    screenedarriving
    Searchable index+0
    • Waiting for the first moment…
    1. 01

      Footage comes in

      Uploaded files, camera exports, screen recordings, meeting recordings, Loom imports.

    2. 02

      Every frame is read

      Motion screening checks every frame and sends the ones that matter to the heavy models.

    3. 03

      It becomes searchable

      People, vehicles, plates, activities and speech are indexed with the file and second they came from.

    4. 04

      A person decides

      The system ranks and points. Every conclusion is a human decision, and it is recorded.

    Measured, not promised

    We publish our numbers with the conditions attached.

    Tested on real CCTV and public benchmarks. Where something isn't measured yet, we say so instead of rounding up.

    0.936

    Vehicle matching (AUC)

    Same vs different vehicle across cameras. 0.951 on our own footage; up from 0.63 on our previous engine.

    Tested on: VeRi-776 public benchmark

    22 / 22

    Plates read exactly

    Barrier CCTV with known plates; cameras close and head-on.

    Tested on: AGH University of Kraków Parking Database

    ≤ 12%

    Speed vs GPS

    Pole camera with a marked stretch of road, against an RTK-GNSS-tracked car.

    Tested on: Intersection Monitoring dataset (CSIC)

    +4.5%

    To screen every frame

    Extra compute for examining every frame instead of sampling.

    Tested on: Measured on our own footage

    How we build

    Decision support, never a verdict.

    Video of people is sensitive, and an answer from footage may end up in front of a manager, a client or a court. These rules are in the product, not in a policy document.

    It never concludes
    “Bag unattended for 11 seconds” is an observation. “Theft” is a word only a person uses.
    Ranked, never hidden
    Uncertain results are shown lower down, not thrown away. You can’t dismiss what you can’t see.
    Empty means unknown
    “Nothing found” is never allowed to mean “not checked”. If a step could not run, the page says so.
    The method is named
    Every result states which models produced it, so anyone can ask how it was made.
    Nothing filed on a guess
    Viddy only creates tasks people confirmed. Everything unclear waits for someone to approve it.
    Every workspace walled off
    Role-based access and database-level isolation between organisations. Credentials are encrypted.

    For builders

    Your AI tools can use it too.

    VidForgeX runs an MCP server, so assistants like Claude and Cursor can search footage, ask about a recording, or pull a person's history — with the same access rules as the app. A REST API covers everything else.

    Read the API docs
    mcp.vidforgex.com11 tools
    find_person_in_video— timestamps where a person appears
    ask_about_session— question one recording
    get_person_progression— a person’s history over time
    get_sop_compliance— how a session followed the SOP
    generate_sop_document— draft an SOP from real work

    Bring one real recording.

    In a 30-minute demo we'll run it through the product with you — footage from a case, or a recording of a process your team repeats every day.

    Your video already knows the answer.