Edit Mind

Edit Mind's core promise is simple. You describe what you want in everyday words, and it finds the moment. Type a phrase like a chef plating a dish or someone laughing at a desk, and semantic video search returns matching segments. It works because every clip gets embedded into a vector database, turning visuals and speech into something a query can compare against.

The indexer transcribes the audio track of each video, so search isn't limited to what's visible. Remember a sentence but not the clip? Search the words. Edit Mind points you to the scene where they were spoken. Transcription is handled by Whisper, a well-known speech recognition model, running locally.

Beyond speech, Edit Mind studies the frames themselves. It runs object and text detection, recognizes faces, and analyzes scenes, so a search for a specific prop, a person, or a setting can surface results that a transcript alone would miss. This mix of signals is what makes the search feel less like a filename filter and more like asking a colleague who watched everything. It just works.

Every part of the pipeline runs on your own computer or server, which is the whole point of the design. Footage never goes to a cloud service, so confidential interviews, unreleased client work, and personal archives stay where you put them. For projects that require everything to stay offline, that constraint is baked in. Not bolted on.

You don't babysit the process. A background service watches your folders for new video files and queues them automatically for analysis. Drop a new shoot into the watched directory and it gets picked up and indexed without you issuing commands, which saves real time when you're adding footage in batches.

For people who'd rather skip Docker, there's a commercial desktop app that packages the same engine into a standard installer for macOS and Windows. It adds direct integration with DaVinci Resolve and Final Cut Pro, so found moments can move into your edit without a round trip through the filesystem. The developer notes the desktop build can also tap Apple silicon GPUs, something the Docker version can't do.

The self-hosted version is open source and free, built on a stack that includes ChromaDB for vector search and PostgreSQL for relational data. Something broken? Read the code. File an issue. Send a fix. The developer has said the free community version isn't going away, and the paid app exists to fund its continued development.