Keyword search
exact matches
A full-text index (SQLite FTS5 with BM25 ranking) over filenames, tags, captions, and metadata. Instant, exact, and the reliable floor. It still works even if the ML models aren’t installed.
Google-Photos-grade video search for your own drives, running fully offline
Your video library,
finally searchable.
Organizamo is a desktop app that catalogs, tags, and searches your local video collection using AI that runs entirely on your own machine. No uploads. No subscriptions to your own footage. Nothing leaves your device.
Most of us have a folder (or a drive, or a NAS) full of video we’ll never find again. Filenames lie. Folders don’t scale. And the tools that actually understand video content want you to upload everything to someone else’s servers.
Organizamo takes a different position. It scans your local video folders, understands what’s in the footage, and lets you search it in plain language: “outdoor interview,” “whiteboard session,” “drone shot over water.” Then it jumps you straight to the scene, even in a video you’ve never watched. All of the analysis happens on your computer.
What sets it apart
-
Searches what’s on screen
Natural-language search across content, scenes, and auto-generated tags, not just the filename.
-
Scene-level results
It returns the exact moment inside a video, not just “this 90-minute file probably has it.”
-
100% local, by design
Every model runs on-device. Your footage is never uploaded, indexed by a third party, or rate-limited.
-
Built for big libraries
Designed for hundreds to thousands of videos, with background processing that stays out of your way.
-
Set-and-forget
Watches your folders and indexes new videos automatically as you add them.
Technology
Real AI. Genuinely local.
Organizamo’s search is tri-modal: every query runs through three independent retrieval engines at once, and the results are fused into a single ranked list. All three run locally: no API call, no cloud inference, no telemetry on what you search for.
Semantic text search
meaning, not letters
A sentence-embedding model (e5-small-v2, 384-dimensional) turns your query and the library’s text into vectors, so “kids playing in the yard” can match a clip captioned “two children running on grass.” It runs locally, with no Python required for the shipping path.
Visual search
what the footage looks like
A CLIP-class image model (CLIP ViT-B/32, 512-dimensional; SigLIP also supported) embeds video keyframes into the same conceptual space as text. Describe a scene in words like “snow-covered mountain” or “person at a podium,” and match it against pixels, even if no one ever tagged it.
How the results come together
The three engines aren’t just concatenated. A fusion layer normalizes and merges their rankings using established information-retrieval algorithms (Reciprocal Rank Fusion, weighted scoring, Borda count, and Condorcet methods), with an adaptive strategy that weights each engine based on the query. Results are de-duplicated and diversified so you get one clean ranked answer instead of three competing lists.
Understanding the video itself
Before any of that can work, Organizamo has to understand the footage:
- Scene detection segments each video into time-aligned units, so search can return a timestamp, not a whole file.
- Object detection (YOLOv8-nano) identifies entities in keyframes.
- Captioning (a BLIP model) writes natural-language scene descriptions that feed the text and semantic indexes.
This runs as background jobs on a priority queue, so indexing a large library never blocks the app.
Where the vectors live
Embeddings are stored in a local vector index (sqlite-vec, cosine similarity, scaling to millions of vectors) with an exact-similarity fallback. The entire index is a file on your disk: portable, inspectable, and yours.
Plenty of products say “AI-powered” and mean “we send your data to an API.” Organizamo means the opposite: the models are downloaded once and executed on your hardware. The trade-off is honest: indexing takes some CPU time up front, and in exchange your library is genuinely private and works with no internet connection at all.
Under the hood: Electron + React, SQLite (FTS5 + sqlite-vec), CLIP/SigLIP and e5 embedding models, YOLOv8, and BLIP, all running on-device.
Not quite ready, but close
Be the first to know when Organizamo launches.
We’re putting the finishing touches on it. Leave your email and we’ll tell you the moment it’s available. No spam, no list-selling, one email when it ships (and maybe one when the beta opens).
You’re on the list.
We’ll be in touch when Organizamo is ready.