What I built
DeskGraph is a local-first computer context graph for the files people choose to authorize. It builds a local SQLite manifest, optionally extracts bounded text from selected scanned documents, and makes that information searchable without requiring an API key, Python, Docker, Ollama, or a cloud upload.
The current build demonstrates explicit native folder authorization, a metadata-only scan, bounded text/PDF/Office extraction, offline Traditional Chinese and English lexical search, conservative project, duplicate, and version signals, a suggestion-only Smart Cleanup Inbox, and a read-only MCP search_files tool. Results explain their matching source, and extracted text is always treated as untrusted data.
How I built it
Rust owns filesystem policy, SQLite state, extraction jobs, graph and retrieval logic, transactions, and MCP. Tauri with React and TypeScript provides the local desktop interface. SQLite FTS5 provides offline lexical retrieval. Apple Vision OCR is a bounded macOS development provider for explicitly selected PNG and JPEG inputs.
Codex, powered by GPT-5.6, was used as a development collaborator to investigate the codebase, split work into safety-bounded vertical slices, implement and review changes under human direction, and run Rust and TypeScript validation gates. It does not receive a user's DeskGraph database or files by default, and it never directly executes filesystem operations in the product.
One concrete safety review reproduced a Unix wrong-inode race between final validation and a pathname rename. Instead of weakening the invariant for a demo, I rejected that production execution adapter and kept Rename and Move preview-only.
Why it matters
Many AI file organizers ask people to trust opaque automation with their entire computer. DeskGraph starts with a smaller, useful promise: give a person and their AI tools a controllable view of explicitly selected local context. It prefers preview, evidence, and refusal over unsafe mutation. That boundary is what can make future organization automation trustworthy.
Current limitations
DeskGraph is a pre-release development build and the demo uses synthetic files. The current usable search is lexical, not vector or hybrid. Cleanup and Rename are preview-only; production Trash, Undo, installers, and cross-platform release validation are not claimed.
Built With
- apple-vision
- codex
- fts5
- gpt-5.6
- mcp
- react
- rust
- sqlite
- tauri
- typescript
