Drive Catalog Sitemap

English | 简体中文

Site Map (Drive Catalog)

The Site Map is ShareOneList's local directory map of every OneDrive and SharePoint document library you actually use. It powers smarter cross-site search and gives the AI assistant a sense of where your files live — without any background crawler, and without ever storing file contents.

• Stored in a local SQLite database (catalog.db) with a full-text index (FTS5, trigram).
• Holds metadata only: paths, names, short descriptions, visit counts. Never file contents.
• Grows by following your real behavior — the map converges to the folders you actually work in.

1. How the map grows

There is deliberately no automatic deep crawl. Registration seeds only the root level, then the map builds itself from three kinds of everyday actions:

Trigger	What gets recorded
You browse a folder	The folder's children are upserted with fresh names/IDs, and the folder's visit count increases. Browsing a folder also refreshes it — renames, moves, and deletions sync naturally the next time you look at it.
A search or AI answer hits a file	The file is recorded together with its full ancestor folder chain — zero extra API calls. Next time a similar question comes up, the catalog already knows that folder.
The AI reads a file's content	The file's visit count increases and its description is filled with what the file was about (only if empty — never overwritten).

Every write-back is logged to a usage ledger, so "how did the AI know about this folder" is always auditable.

Optional manual deep index. Want the full map right now? Open the Site Map page and run a deep index for a single drive, up to a depth you choose (default 5, UI cap 10). It shows a progress bar, can be cancelled, and is never triggered automatically.

2. What the Site Map page shows

Sidebar → Site Map:

• Registered drives — account, site/name, status badge (seeding / ready / failed), node count, last-used time.
• Tree view — browse the accumulated directory tree lazily; visited hot spots get visual heat signals (🔥×N), unexplored areas are simply blank.
• Search mode — full-text search over the map. The index is trigram-based, so any substring of 3+ characters matches, including Chinese; shorter queries fall back to plain matching. Results are ranked by visit count and path depth.
• Accumulation stats — e.g. "N folders learned from browsing this week", straight from the usage ledger.
• Per-drive actions — deep index (with progress + cancel), re-index, unregister drive.

All queries are strictly account-isolated: you only ever see data of currently logged-in accounts, and Global / China (21Vianet) sessions never mix.

3. How the AI uses the map

When you Ask AI a question:

1. The question's keywords are queried against the catalog first.
2. The most relevant drives (weighted by visit counts, top 4) are searched in Microsoft Graph with expanded synonym terms, and likely folder paths are handed to the model as "possible locations".
3. Without catalog hits, the assistant falls back to plain OneDrive search — exactly the behavior you'd get without the map.
4. The hits (with ancestor chains) and any file reads are written back, so every answer makes the map a little better.

This is also what fixes "Graph search doesn't understand my long Chinese sentence": the catalog matches folder names by substring first, then only the promising drives get the Graph query.

4. Data safety

• catalog.db is derived data: it contains nothing you couldn't rebuild by browsing. If it's ever corrupted, it is deleted and rebuilt; a fresh install starts from root seeds and regrows as you work.
• Only drives you can already see and browse are registered — the catalog never asks for extra permissions.
• Everything stays on this machine. Unregistering a drive removes its nodes immediately.

See also

• AI-Assistant — the consumer of this map
• AI-Memory — what the assistant remembers about you