remnus

Everything created, updated or deleted since a cursor (from the workspace digest/map or a previous call) or an ISO time — pages, databases, rows and deletions, oldest first. Omit both to bootstrap. Always keep nextCursor for the next call: this is how to catch up instead of re-reading the tree. An entry stamped in the cursor's own second may appear once more — dedupe by id.

Columns (id, name, type, options) and saved views of a database, without rows. Check it before writing rows or changing views.

Read a page, database row or dashboard by id (type auto-detected; a dashboard comes back as its block spec). mode: "outline" returns headings + first line per section — a cheap skim for a long page (the map shows body sizes) before a full read.

Read several pages/rows by id in one call, for a known, possibly mixed id list. A bad id fails only its own entry — check each "ok". For rows of one database use query_database with filters/fields instead: one query, not N lookups.

A page's neighborhood, titles and ids only: parent, children, outgoing links (@-links and child blocks), backlinks, row siblings, and the repo files it rests on. With resource (a file path) instead: the pages resting on that file — read them before changing it.

This connection's state: workspace, scope, token expiry and renewal. Call it when Remnus tools are missing, a write is refused or a call fails with 401.

Workspace members with roles and join dates.

List sidebar items (pages and databases), optionally under one parent; paginated. For orientation the workspace digest/map is cheaper.

Build a task-specific, token-budgeted context pack from the most relevant pages: lexical rank, human-reviewed knowledge preferred, stale/deprecated concepts penalized, plus the top hits' link neighbors. Use it before multi-page product or coding work instead of many search/get_page calls.

Agent activity log for this workspace, filterable by tool, status and date range.

Rows (and the matching schema) of a database, paginated. Row bodies are NOT included unless "content" is in fields. Use filters to narrow and fields to project only the columns you need — much cheaper on wide tables.

Case- and accent-insensitive search over titles and bodies of pages, databases and rows, best matches first. Use it to locate an item by text when the workspace map does not already give you its id.