Point Cabinet at your messiest folders. It reads every document — PDFs, Word files, text — and files them into smart virtual folders organised by what they're actually about, using the LLM of your choice. Your real folders stay exactly as they are.
What it does
Categories are a browsable tree over your files, not a reorganisation of them. Nothing is moved, renamed or touched on disk — ever.
Text is extracted from PDFs, Word documents and plain text, so filing decisions are based on content, not just filenames.
Works with Anthropic, OpenAI and Gemini — or point it at a local Ollama model and nothing leaves your machine at all.
Re-file something by hand and it stays filed. Re-running the AI only ever revisits its own decisions, never yours.
Files that appear, change or vanish are reconciled on every scan — changed documents are re-read, missing ones flagged, returning ones welcomed back.
Category tree, sortable file table and details panel in a fast three-pane desktop app, with live progress while filing runs.
How it works
Choose folders to watch. Cabinet inventories what's there.
Each document's text is pulled out for reading.
The LLM builds and refines a category tree, then files each document.
Explore by topic instead of by wherever things happened to land.
Principles
Read-only by design. Cabinet never writes to your folders. The worst it can be is wrong — never destructive.
Local-first. The app, database and extracted text live on your machine. With a local model, so does everything else.
Human override, always. The AI proposes; you dispose. Manual filing outranks the model permanently.