Semantra is an open-source desktop app that brings full semantic search to your own files, entirely on your device. Published on GitHub by developer Dylan Freedman (freedmand), it indexes images, documents, videos, audio and spreadsheets locally and lets you find them by meaning rather than by exact keywords or filenames. The repository is small but active, with 19 stars and development continuing into 2026.
Semantra is built around Google’s open-source EmbeddingGemma 2 model, which produces embeddings across text, images, audio and video and runs completely on your own hardware. On Apple Silicon the model runs through MLX; on Windows, Linux and Intel Macs it runs through ONNX Runtime, using GPU acceleration via WebGPU where available. You simply add files — images, text and PDF documents, videos, audio recordings or CSVs — and Semantra chunks, embeds and indexes them in a local vector store. Searches then combine semantic similarity with classic keyword matching, so a query like “sunset over the harbour” can surface a photo you never tagged or renamed. Because everything is computed and stored locally, no documents ever leave your computer.
Semantra is Free. The source code is published on GitHub and ready-made installers — a macOS disk image, a Windows setup executable and Linux AppImage and .deb packages — are provided as free downloads. There is no paid tier and no hosted service.
Semantra suits anyone sitting on large personal collections of photos, recordings, videos or documents — researchers, journalists, creators and archivists — who wants meaning-based search across mixed media without handing their files to a cloud service.