Loading content...
A local-first filesystem agent that understands files, protects developer workspaces and makes every action reversible.
A local-first filesystem agent built around protected boundaries, hybrid retrieval, observable actions and one-click rollback.
People remember what a file contains, not whether it is in Downloads, Desktop or a nested project folder. A useful agent could understand those files, rename them and place them somewhere sensible. The same capability is dangerous when applied to a developer machine: a mistaken move can break a repository, destroy expected paths or organize generated files that should never be touched. TidyOS therefore treated safety as the product foundation rather than a later permission dialog.
The application extracts content from PDFs, DOCX, text and images, then combines semantic retrieval with SQLite FTS5 so users can search by remembered meaning and exact terms. Watchdog monitors approved intake folders, but stabilization checks wait for downloads to finish before processing. Protected-workspace detection identifies Git, Node, Python and other structured project markers. The agent proposes names and destinations within approved roots, records file operations in a local SQLite audit ledger and supports reversal. LangGraph coordinates understanding and actions, while deterministic policies decide what the agent must not touch.
Keep files and the audit ledger local rather than requiring cloud upload.
Combine vector retrieval with SQLite FTS5 so semantic and exact-term search support each other.
Detect codebases and structured workspaces before allowing organization actions.
Make moves reversible and observable instead of granting the agent unrestricted filesystem autonomy.
Stabilize new files before processing to avoid acting on incomplete downloads.
Test the RAG retrieval pipeline by asking specific questions about this case study: