An in-house AI that turns a 256 TB creative archive into a conversation
A self-built, self-hosted system for a motion & post-production studio. Ask a question in plain language — about any file, project, task, or review — and get an instant, sourced answer.
The impact: a five-system scavenger hunt becomes one plain-language question — and the same assistant surfaces untracked terabytes that were quietly costing storage. Built, shipped, and in daily use at a working studio.
At a glance: ~256 TB · ~2.17 million files · 5 disconnected systems unified · 4 ways to search · reachable from the studio's own chat · 100% self-hosted & private.
🧩 The problem
A working creative studio leaves an enormous, messy trail of data:
- ~256 TB of footage, projects, and exports — ~2.17 million files spread across a local NAS and a cloud workspace.
- Critical information lives in five separate systems that don't talk to each other: the file servers, the cloud drive, the task manager, the review platform, and shared documents.
- Finding "the final 4×5 export of a given campaign" or "which projects are quietly eating space on the active server" meant digging through folders and apps by hand — slow, error-prone, and dependent on who happened to remember what.
In industry terms, there was no Media Asset Management (MAM) layer tying it together — assets, proxies, deliverables, and review notes sat in separate systems, invisible to the people doing transfer, transcode, delivery, and archiving every day.
The knowledge was all there. It just wasn't askable.
💬 What it does
A single chat window that understands the whole studio. People ask in plain English — or by voice — and it does the lookup, the reasoning, and the presentation for them.
🔎 Find anything, four ways
- 🏷️ By name or path — exact lookups across millions of files, instantly.
- 🧠 By meaning — semantic search ("the moody slow-mo perfume spots"), not just keyword matching.
- 👁️ By what's in the frame — an AI "watches" sampled video frames, so a shot can be found by its visual content even when nothing in the filename describes it. Or drop in a reference image and it finds footage that looks like it.
- 🗣️ By what's said in it — spoken audio is transcribed, so a clip can be found by a line of dialogue or a spoken idea, not just its filename.