Storage

Every analysis is persisted in the browser, using IndexedDB. There's no database to run, no port to configure, no daemon to keep alive — it works the moment the extension is loaded, which is also what makes it Chrome-Web-Store-ready.

What's stored

• Repos — each analysis payload, keyed by a stable hash of the repo id.
• Decisions & collections — your Adopt / Trial / Hold / Reject calls (with notes and dates) and the collections you've grouped repos into.
• Evaluations & notes — your rubric and per-repo scores, plus any personal notes.
• Graph — the nodes and edges behind the Connections tab.
• Scan history — a local cache of recent scans, so reopening a repo is instant and costs no AI call.

Search across your library is a fast client-side BM25 ranker (the algorithm behind real search engines): rarer query words count for more, and a hit in a high-signal field like the category or capabilities outranks a passing mention buried in the summary. No embeddings, no vector database, no network.

Manage your library

The Library tab is mission control for everything you've scanned:

• Sort & filter — by best fit, health, stars, recency, or name; filter by capability; search as you type.
• Bulk delete — hit Select, tick any number of cards (or Select all), and remove them in one confirmed action. Esc or Done leaves selection mode.
• Stats bar — a one-glance read on the shape of your library (how many strong/solid/care/risky, average health).

Back it up — and take it with you

Because it's your data, you can move it between machines:

• Library → Export writes your whole library — analyzed repos, your decisions and collections, the Connections graph, and the scan cache — to one portable JSON file. Import restores it (merge or replace). Backups are validated and size-bounded on import, so a bad file fails safe instead of corrupting your library.
• Options → Back up your settings exports your preferences — theme, voice, model picks, per-part routing. It is allowlist-driven by design: your API keys and tokens are never written to the file (and are stripped out even if someone hand-edits one in).

Importing an old VelesDB library

Earlier builds of RepoLens saved to a local VelesDB server. If you have one, you can pull that library into the new in-browser store in one pass:

1. Start your VelesDB server.
2. Open Options → Import from VelesDB.
3. Confirm the URL (defaults to http://localhost:9090) and click Import library.

It reads every saved repo and copies it across. The import is idempotent — re-running overwrites by id, never duplicates. The Connections graph isn't migrated; it rebuilds naturally as you use the app.

Repos saved by older builds may show as Unrated — their fit data was never persisted. Re-scanning a repo fills in its fit chip.

A seam for later

Persistence sits behind a single storage interface. Today the only backend is IndexedDB; a future sync or shared-store backend can slot in behind the same interface without touching the rest of the app.