Repository navigation
First release. Metachlorian watches your footage once, on your own computer, and builds a shot-by-shot index that you and AI agents can search: by what's in a shot, what's said, how it was filmed and whether you're cleared to use it. Download the desktop app below for Windows, macOS (Apple Silicon) or Linux, or run the server with Docker (ghcr.io/jeremysnr/metachlorian:0.1.0).
Added
- The desktop app installs everything it needs. The Windows installer, the macOS disk image and the Linux AppImage now include their own Python, the Metachlorian engine, FFmpeg and the web app, so nothing else has to be installed first.
- The first time it opens, the desktop app downloads the analysis models (about 1.7 GB, once) and shows its progress. If the download fails it says so and offers Try again, keeping anything already downloaded.
Fixed
- Metachlorian now runs on Windows. It previously stopped at start-up there, because it used a file lock that only exists on macOS and Linux.
- Semantic search and zero-shot labels now work on Apple Silicon Macs. Embedding shots failed when analysis used Apple's Core ML accelerator; it now runs on the CPU unless
METACHLORIAN_COREML=1is set. - The macOS desktop download no longer opens with "Metachlorian is damaged and can't be opened". The app is now ad-hoc signed, so macOS shows its usual unidentified-developer prompt and System Settings → Privacy & Security → Open Anyway works. It is still not signed with an Apple Developer ID or notarised.