Recall offers local multimodal semantic search for files using Gemini Embedding 2 and ChromaDB. The open-source tool indexes images, audio, video, PDFs, and text locally, enabling natural language queries like "team dinner" to surface relevant photos without metadata. Built entirely on your machine, it only sends files to Google's API for embedding, keeping your data private.

Setup requires cloning the GitHub repository and running an animated wizard that handles API key validation, folder selection, and indexing. The tool includes a Raycast extension for instant visual search from your launcher. Cross-modal search works out of the box—no tagging, renaming, or manual metadata entry required. Files are SHA-256 deduplicated to prevent redundant indexing.

The Python API supports embedding single files or entire directories, with natural language search returning ranked results by similarity. The Raycast extension provides both grid UI with thumbnails and headless instant file opening. Supported formats include common image types (.png, .jpg, .webp), audio (.mp3, .wav), video (.mp4, .mov), documents (.pdf, .txt), and various code files. Configuration options include setting the data directory, API key, and supported file types through environment variables.