Lists — a list of things with their data, kept in sync, tagged, embedded and ranked

A list holds items: a stable key and the source's JSON, verbatim. You get one from a CSV or JSON upload, from a web page (a table or a list beneath a heading, or a JSON endpoint), or — with the X connection — from an X list. Items are tagged with your account's tags, can be embedded for semantic search and "find similar", and can be ranked by a decision model against a rubric you write.

Turn the block on

Under Blocks, enable Lists. The card's Open link goes to the default project's lists at /a/<account>/p/<project>/lists.

Create a list and add a source

New list asks for a name, a slug, the field that becomes each item's title, and how long an item that leaves the list is kept (the account default is 90 days; indefinitely never purges by age). Then Add source: paste a CSV or JSON, name an uploaded file from the media library, or give a page URL with the heading to read beneath. Sync now runs it; the source line shows what came in, what changed, what left the list, and any rows that do not match your schema hint (they are stored anyway — the hint reports, it never rejects).

A source marked authoritative removes the items it stops returning; an upload is additive unless you say otherwise. A list has one authoritative source at most.

Tags, search, similar

Select items and add tags (tier:1, topic:ai) — the same vocabulary as your media and threads. Filter by text, by tags, or — once embeddings are on — by meaning ("people who write about agent operations"). Each item's page offers find similar.

Promote an item to a person

An item stays data until a person promotes it: the item is linked to a party (by e-mail, or by handle and host) and shows the party card from then on. When an item's e-mail already belongs to a party in your account, the page shows matches … in People — the link is made by your click, never by the sync.

Rank with a decision model

Write a rubric — one sentence of context and the questions: a score with ordered levels for the headline number, and optionally a choice and a yes/no noul. Estimate the cost shows the credits; Run it ranks every item and the table orders by the run. Treat the order as a draft.

Agents

Through MCP, agents read lists (list.query, list.get_item, list.similar) and propose writes (list.tag_by_filter, list.promote, list.rank, list.sync). A proposal shows up on the list page with the agent's reason; confirm it or reject it — nothing is written until you do.