What hardware does it need?
A machine with a reasonable NVIDIA GPU. Indexing is the expensive part and it happens once per asset; searching afterward is cheap. We will size it against your actual library rather than quoting a spec sheet. The real answer depends on how many hours you have and how quickly you need them indexed.
Do I have to move my files?
No. It can index media where it already sits, and metadata sidecars in standard fields are available on request. If you would rather it held the library itself, it can copy new media into a managed store as it ingests. Archiving masters to cold storage and restoring them without relinking is on the roadmap.
How good is the search, really?
Good enough that people stop scrubbing, and not good enough to be left unsupervised. It returns ranked, timecoded candidates that show why each one matched, so you can tell at a glance whether it found the right moment. On our own scored tests, on our own test libraries, it puts the right clip first the large majority of the time. We would rather show you that number on your footage than quote ours.
Does it replace a logger?
No. It removes the re-watching, not the judgment. Somebody still decides what is good; Mokuji tells you what is there.
What happens when it gets something wrong?
You fix it, and the fix sticks. Captions are fully editable, identities can be renamed in place, and a correction propagates rather than being overwritten on the next pass. Recognition is a strong hint for a person to check, not an assertion. Anything with consequences attached should have a person confirming it.
Does it work with an archive that has no metadata at all?
That is the case it was built for. It reads the footage itself, so filenames like A001_C047.mov and a spreadsheet somebody abandoned in 2019 are not an obstacle. If anything, the worse your existing metadata, the larger the difference.
Can it run with no internet connection?
It is built to. Mokuji is built to run with no internet connection once installed; models are staged locally, and Mokuji's own services report nothing back. A formal offline acceptance test is not yet part of each release, so we run one with you before you rely on it.
What does it cost to leave?
Your media is untouched on your own storage, and your selects, markers and metadata export to open formats: timeline XML, CSV and sidecars, with a documented API for the rest. That is deliberate: a product that has to trap you is a product that expects you to want out.
Is my footage used to train anything?
No. It runs on your hardware and nothing is ever sent to us. In the standard install nothing leaves your network. Cloud processing is optional and off by default: if an administrator turns it on for a catalog, that catalog's GPU work goes, encrypted end to end, to a cloud GPU provider you choose, and is dropped when each job ends. This is the question we get asked most, and the architecture is the answer rather than the policy being the answer.
Ask us the one that is not here.
Particularly the awkward one. We would rather answer it now than have it surface after a deployment.