Command-line semantic search for your offline photo stash

What it does rclip indexes your local image collection and lets you query it with natural language or reference images straight from the terminal. It uses OpenCLIP’s ViT-B/32 model to score matches, then returns ranked file paths. Think of it as grep for visual meaning instead of text strings.

The interesting bit The query algebra is the standout feature: you can add and subtract text and image queries on the fly—search for “horse + stripes” or “racing car - sports car + snow” to nudge results without retraining anything. The tool also incrementally updates its index, skipping unchanged files on later runs so you are not punished for growing your library.

Key highlights

• Indexes and searches local images using OpenCLIP’s ViT-B/32 model entirely on your own hardware.
• Supports mixed-modality queries: text strings, local image paths, URLs, and arithmetic combinations using + and -.
• Incremental indexing: subsequent scans only process added or deleted images, and --no-indexing can bypass scanning entirely if nothing changed.
• Packaged for Linux (snap, AppImage, pip), macOS Apple Silicon (Homebrew, pip), and Windows (MSI installer).
• Optional inline image previews in a handful of supported terminals via --preview.

Caveats

• macOS support is Apple Silicon only; Intel Macs are left out.
• Initial indexing is CPU-bound and can take hours on large libraries or low-end hardware—the author reports roughly a day for 73,000 images on an old Celeron NAS.
• The built-in image preview relies on a specific allow-list of terminals—iTerm2, Konsole, WezTerm, Mintty, and mlterm—so most other emulators will need external viewers piped manually.

Verdict Photographers, digital hoarders, and anyone managing large local image libraries who lives in the terminal should take a look. If you already have a polished GUI DAM with AI search, or your collection lives entirely in the cloud, this probably will not displace your workflow.