Topic analysis

Recall – local multimodal semantic search for your files

Local multimodal memory with semantic search. Embed images, audio, video, PDFs, and text into a local vector database — then find anything with a natural language query. A text search for "team dinner" surfaces the photos, even though the photos have no text metadata. Comes with an animated setup wizard and a Raycast extension for instant visual search. Powered by Gemini Embedding 2 (768-dim, free tier) and ChromaDB stored entirely on your machine. Cross-modal search works out of the box. No tagging, no renaming, no metadata required. That's it. The animated setup wizard handles everything end-to-end: Manual key setup: cp .env.example .env and add GEMINI_API_KEY=your_key ingest_file returns {"status": "embedded" | "skipped" | "error", ...} . Files are SHA-256 deduplicated — re-ingesting the same file is a no-op. Visual grid search with image thumbnails, right from your launcher. Open Raycast and search Memory Search . On first launch, go to Preferences and set: Tip: Run python setup_wizard.py first — it prints these values pre-filled for you. All vectors are stored locally in data/chromadb/ . The only outbound traffic is embedding API calls to Google — your files never leave your machine.