Newsroom · Enterprise
DeepFile
Filename search fails the moment you cannot remember the filename. At the scale a media library actually runs at, it fails all the time — and for video it never worked in the first place.
Overview
A news or media organisation accumulates an enormous amount of material that is not the published story. Contracts and rights letters, releases, scripts, transcripts, research packs, stills, contact sheets, scanned correspondence, decades of it, across systems that were each somebody's good idea at the time.
None of it is findable by name. The rights letter is scan_0043.pdf, the transcript is in a folder called stuff, and the photograph you need is somewhere in ninety thousand images. That is why the practical answer to "do we have the paperwork for this" is so often a person spending an afternoon looking.
DeepFile reads the contents — documents, spreadsheets, PDFs, scans, photographs, and text recognised inside images — and builds an index of what each file is about. You then ask for it the way you would describe it to a colleague.
Video is indexed too, and this is the part that surprises people. Footage becomes searchable by the people in it — faces are identified and indexed across the whole library — and by describing what happens in it. People and crowds are tracked through the footage, so a result is a stretch of video with a beginning and an end. That is what turns "find the clip where the minister is on the steps" from an afternoon into a query.
Finding it is only half the job, and it is the half that scales worst. The agent is the other half: from any result you go straight into asking what the document says, what it commits you to, and how it relates to the others it came back with. At archive scale that is the difference between retrieval and an answer.
Capabilities
What DeepFile does
Search by meaning
Describe the document instead of naming it. Retrieval runs over content, not over a filename substring, which is the only thing that works once the archive is past browsing size.
An agent that reads the result
From any result, move straight into asking what the document says rather than opening it and reading. The agent stays on that material, so the answer is grounded in your file and not in general knowledge.
Context across the archive
Ask about a set of files rather than one. The agent works across what the search returned, so a question that spans a contract, an amendment and an email thread gets one answer.
Reads what is inside
Documents, spreadsheets, PDFs, scans, photographs and video, including text recognised inside images — which is where most of an older archive actually lives.
Video, searchable by who and what
Footage is searchable by the people in it and by describing what happens, rather than by whatever the filename claims. People and crowds are tracked through the shot.
Every platform, self-hosted or cloud
Android, iOS, macOS, Windows and web from one codebase, pointed either at your own backend or at Manotr cloud.
The part nobody solves
Searching video by who is in it
A video archive is the least searchable thing an organisation owns. There is no text to index and no filename worth trusting, so in practice the answer to "do we have footage of this" is a person scrubbing tape. DeepFile makes footage answerable the same way a document is.
Searching video by who is in it
VideoSearch by person
Faces are identified and indexed across the whole library, so footage of an individual can be gathered from material shot years and formats apart. No accuracy figures on a marketing page — we will run it against your own footage instead.
Search by description
Describe what happens and get the footage back. Nobody has to have logged it by hand first, which is why most archives are unsearchable in the first place.
People and crowds, tracked
Individuals and crowds are followed through the footage, so a match comes back as a stretch of video with a beginning and an end rather than an isolated still to go hunting around.
Several models, one answer
More than one AI model works on the same material inside our pipeline, and the results are combined. That is what lets one plain-language question resolve across a video library at all.
Who it is for
Built for these situations
News and media archives
Publishers and broadcasters holding decades of rights paperwork, releases, transcripts, research and stills that is technically stored and practically unreachable.
Rights and compliance teams
The people who have to answer "are we cleared to use this" against material nobody has catalogued since it was filed.
Enterprises with sensitive files
Teams that need semantic document search but cannot place their document store in a third-party index.
Deployment
- Self-hosted on infrastructure you control, or Manotr cloud.
- Native apps for Android, iOS, macOS and Windows, plus web.
- Runs against a hosted model API or a local model where nothing may leave the network.
- Batch ingest as jobs — an archive is not a file-at-a-time activity.
Questions
Frequently asked
Does it handle video?
Yes. Video is searchable both by the people in it and by describing what happens in it, with people and crowds tracked through the footage. Documents, images and video sit in one index, so a single question can cross all three.
How is this different from Media Pipeline?
DeepFile is about finding and understanding what is already in your library — search, questions, people, subjects. Media Pipeline is about preparing material to ship: remastering, rights matching, in-player context, moderation and delivery cost. Media organisations usually run both, against the same archive.
Is anything uploaded to a cloud service?
Only if you choose the cloud option. Self-hosted, indexing and search both run inside your own infrastructure.
How large an archive can it handle?
It is built for archives past the point where browsing works. Ingest happens as batch jobs rather than file by file, and scale is something we size against your actual volume during the engagement.
Can I ask questions about a document rather than just find it?
Yes, and that is the point of it. Every search result opens straight into a conversation about that file, and the agent can work across a set of results at once.
Interested in DeepFile?
Every deployment is customised to the organisation running it. Tell us about your environment and your constraints, and we will tell you honestly whether this is the right fit.