An AI assistant for the footage on your NAS

The NAS solved the wrong half of the problem. It gave you one address for everything you have ever shot, which is useful, and then it left you with forty terabytes you can browse and cannot search. A shared folder is not a library. It is a very large filing cabinet with no index card in any drawer.

The verdict up front: the fix is not more storage, better folder discipline, or a naming convention retrofit. It is a description layer, computed once, written next to the footage, and searchable from your desk.

What a NAS knows about your footage

A NAS (network attached storage: a Synology, a QNAP, a rack of drives with a network port) is a file server. It knows filenames, folder paths, sizes, and dates. That is the entire set. Its built-in search is a filename search, so it can find A001C012.MOV in a second and can never find "the wide of the harbour at dusk", because nothing on the box has ever looked at a frame.

This gets worse with scale in a way people underestimate. At two terabytes you remember roughly where things are, so folders work. At forty terabytes across eleven years and four camera systems, the folder tree encodes exactly one fact per clip: when you shot it. Every other question you will ever ask is about content, and content is the one thing the tree does not hold.

The result is the pile with a power connector, only tidier and more expensive. The economics are the same ones that keep old drives unopened, and they are worked through in your back catalog is an asset.

What an index changes

An index is a searchable table of facts about every clip, built by reading each clip once. Once it exists, the questions change shape. "Do we have any storm footage from the north shore" stops being a two-hour scrub and becomes a filter that returns eleven clips with timecodes. "Which interviews mention the 2021 permit" becomes a full-text search across transcripts. "Where else does this person appear" becomes one click, because the same face was grouped across the whole volume when the pass ran.

None of that requires new footage or a migration. It requires each clip to be read once, and for the result to be written somewhere the next machine can find it.

Where ClipLogger fits, and where it does not

Be clear about the architecture, because this is where NAS tools usually oversell.

ClipLogger does not install on your NAS. There is no Synology package and no QNAP package. It is a native macOS app, Apple silicon, macOS 15 or later, and it works on footage wherever that footage is mounted. Mount the share on your Mac, point the app at the folder, and it treats a network volume the same as a directly attached drive. Nothing is imported and nothing moves: files stay in the ordinary folders you already have.

What it leaves on the NAS is the part that matters. Each managed volume gets a .cliplogger folder holding the on-volume index, project documents, and derivative caches, and every confirmed decision is also written as a readable JSON sidecar beside the clip it describes. That makes the volume self-describing. Mount it on a different Mac in three years and the description is right there next to the media, which is the only arrangement that has ever survived a vendor going away. The argument for that is metadata should outlive the app.

The honest constraints of shared storage

Two editors can point at the same NAS folder at the same time, with one real limitation: the on-volume index takes one writer at a time. A lock file carries the holder's identity and a heartbeat, and a second machine that cannot get the lease works against its own local cache instead of risking a corrupted index. Project documents are built to merge, and confirmed metadata meets in the sidecars, so nobody loses work. But the writer lease is a queue, and you should know it exists before you plan around it.

The other constraint is physics. Analysis reads frames, and frames come over your network. A gigabit link is a real ceiling on how fast a first pass can run against a share. The practical answer is to let it run overnight rather than to fight it, or to run the pass against the copy that is still on your local SSD before it moves to the NAS. On a volume backed by a JuiceMount server, the computed AI is shared through the server, so a second Mac reuses it instead of reading everything again.

Backfilling eleven years without a budget line

The pass over an existing archive is the part that used to be impossible, because it was weeks of unpaid watching. Now it is a scheduling problem.

The local engine is free, needs no account, and keeps every frame on the Mac. That is the right lane for an archive nobody is waiting on: point it at a share, let it grind, come back tomorrow. The paid cloud lane, Rush, is for when a person is waiting on the answer, and it is flat at one credit per clip with the cost shown before the batch runs. Choosing between them per batch, rather than by principle, is covered in on-device vs cloud AI is routing, not religion.

Whichever lane runs, nothing the model produces becomes the record on its own. Proposals stage for review and a person confirms them, for the reason laid out in AI proposes, you confirm: a wrong value in a forty-terabyte index is worse than an empty one, because it teaches everybody to stop trusting the index. What you can realistically expect the models to get right is broken down in analyzing video clips with AI, and the fields worth filling are in a metadata schema for logging video. The category itself, if you are new to it, is what an AI video logger actually does.

Start with one share, not the whole box

Do not index eleven years on the first weekend. Pick the single share with the highest expected value, usually the one tied to your best client or your most licensable subject, and describe that one. Define three fields you would filter on. Run one pass. Review it. Then decide whether the rest of the box earns the same treatment, with evidence instead of optimism.

A NAS gives you one place to keep everything, and a description layer is the only thing that turns that into one place to find anything.