Clusters

A public beta through BRAT, alongside Model Hub and Books & Shelves. First, in order, what the first real index (a shelf of 35 books, 2026-10-05) showed needs fixing. Plan: docs/plans/2026-10-05-first-use-fixes.md.

• The Collections tab catches up after a run. After Index, Find topics or Survey, a collection's row keeps its old state until settings are closed and opened again. The row updates when its job ends.
• A run in progress shows on its collection's row and in its dialog, with its progress, after the job dialog is closed. Today the only trace is Open jobs.
• Books that couldn't be split into passages are named on the collection's row and at the top of the next dry run, with the fix in one line. Today the only sign is a line in the end-of-run report.
• Size on disk. A collection's row and its dialog say how much its index takes on disk, beside its cards and topics.
• The Index dialog in two layers: the counts, the cost and the Index button first; what is split into passages, what is read again and why, where the index lives and the undo line under a fold. When the run ends, a short summary (passages read, cards made, cost, what failed or was left out) with the full report under a fold, and Done as the only action.
• Dry-run numbers that read as a contradiction ("3 books to split, 2 passages to read"): the dialog says that the passages of books not split yet are counted once they are, so the count will grow.
• "Passages", not "units", in the Index dialog and the run summaries. The guide already says passages.
• Converting AZW3 and MOBI books becomes a setting with a default, asked once and remembered, in place of a question before each run.
• Actions where they act: Preview a collection's passages moves into the Index dialog, and the rest of the command list is reviewed one by one for a place in settings or a dialog.
• An icon per source kind on the Collections tab: books, notes, Bases.
• Setup says when an update to Model Hub or Books & Shelves is available, from Obsidian's own check.
• Spend per run on the Collections tab. Runs record their cost.
• Keep a collection up to date on its own: a switch per collection, off by default, that reads only what changed, within a spend limit per day.
• Cheaper, Balanced or Better, per run: one pick on the Index dialog mapped onto Model Hub's provider order, the extract tier and reasoning, for that run only.
• Screenshots and a short demo for the README and the guide
• A first collection end to end on real notes, with the costs in Model Hub's ledger checked against the dry runs
• Tag 0.1.0; a beta thread on the Obsidian forum

• Submission to the Obsidian community directory, and fixes from its review
• The name checked against "Cluster", which is already in the directory
• Model Hub's skills folder through its API, in place of reading Model Hub's settings
• Lenses for notes and Bases. The built-in lenses are shaped for books; notes, journals and the rows of a Base want their own. A design session first: what a note's card is, which fields come from its properties

Needs Model Hub 1.0, which brings the Hub's tool registry and its MCP server. Clusters registers what it can do with the Hub and runs no server of its own.

• Ask a collection a question from the Hub's panel or from Claude Code, Claude Desktop or Codex, and get the passages back with the book and place each comes from
• Follow a passage to its neighbours in other books, and to the places that disagree with it, from the same agent
• Read a card or a topic note as the agent needs it, so a write-up drafted outside Obsidian quotes your index, not its memory