Education Voucher Check: Check your eligibility Try now

Try now
DE|EN

DataSquirrel

DataSquirrel is self-hosted file intelligence: it indexes the storage you point it at — local folders, NAS mounts and Google Drive — and then lets you find anything in it, without a single byte leaving your machine.

Point it at a folder and it crawls, hashes and catalogues every file it finds. From there, NutFinder searches by name and metadata and narrows results by folder, kind, file type, size or date; a duplicate report shows what you are storing more than once; a storage breakdown shows where the disk actually went. On request — never automatically — a local LLM reads and summarizes a file, a selection or a whole folder, and those summaries become searchable alongside filenames. A chat assistant in the sidebar answers questions about your files from those summaries, cites the files it used so you can click straight through to them, and can run a filtered search itself: ask for your largest videos from this year and it works out the filters. When a file changes, its summary is flagged as outdated rather than silently rewritten; when a file disappears, DataSquirrel tells you instead of handing you a dead path.

Everything runs in Docker on your own hardware: a FastAPI service and a Streamlit interface over PostgreSQL 16 with full-text search and pgvector, with Ollama serving the model. No cloud service, no account, no telemetry. One codebase runs as a four-container install on a laptop or, by configuration alone, adds GPU inference, scheduled scans and monitoring — topology is a deployment choice, never a separate code path.

Case Study

Personal and household storage sprawls across laptops, external drives, a NAS and cloud accounts — and the tools that search it well are the ones that require uploading it first. DataSquirrel's goal is to give people that same quality of search over storage they control: filename and metadata search, on-demand content summaries, and a chat assistant that answers with citations, all on hardware they own, with no account and no data leaving the machine. The second goal is architectural — one codebase that installs as four containers on a laptop and scales to GPU inference, scheduled scans and monitoring purely by deployment configuration, never by branching code.

The Team

  • Beatrice

    Bea built the structured logging, the PostgreSQL integration test harness and the duplicate review view.

  • Arsalan

    Arsalan built the backend spine — the FastAPI service, the ingestion pipeline and the orchestration seam.

  • Saschsa

    Sascha built the interface: the app shell, the Dashboard, NutFinder itself, plus the deduplication engine and the Google Drive connector.

  • Niall

    Niall led the architecture and the content-intelligence layer — the database schema, the full-text search behind NutFinder, and the chat assistant that cites the files it answers from.

Next Project

Don't miss anything.
Subscribe to our newsletter.

I would like to receive email updates from neue fische. This decision can be revoked at any time. Information on how we handle your data can be found in our privacy policy.