SiftFind, understand, and organize your files with private local AI

Asking for a change does not carry it out. Sift first shows the change as a before-and-after plan, and you can approve it, decline it, switch individual rows off, edit where a file is going, or save the plan and come back to it with /apply. When you organize a folder, the proposed tree is drawn as each file is sorted, so you can watch where things are going to land. If you apply a batch and then change your mind, /undo puts every file back where it was.

Three local models do three different jobs: one answers questions and organizes folders, one turns your files into the embeddings that make meaning-based search work, and one reads what is inside images. Run /model to see which model is currently filling each role. Sift uses these by default:

ollama pull qwen3:4b               # answers questions, organizes folders
ollama pull qwen3-embedding:0.6b   # meaning-based search
ollama pull qwen2.5vl:3b           # reads text and content inside images

Sift lists every compatible model you already have in Ollama, so you can switch to a different one with /model NAME or from the browser app. Reading images takes a few seconds per file, which is far slower than reading text, so it only happens when you ask for it with /scan --images.

Built for agents too

Every Sift action also runs as a single command: a normal exit code, a plain-language answer, and a plan another program can read. An agent can draft a plan and save it as JSON, a person can review it, and Sift can apply it later, or an agent can call these commands directly as part of its own workflow.

sift status ~/Downloads
sift ask "Where is my March invoice?" --folder ~/Downloads
sift plan "move all PDFs into Papers" --folder ~/Downloads --save papers.json
sift apply papers.json
sift organize ~/Downloads --save organize.json
sift reveal ~/Downloads/invoice.pdf
sift undo