CoChat

Read a CoChat conversation's transcript by id (from chats_list). Returns the active-branch messages as [{role, content}], oldest first. Use this to pull context or a decision from an earlier chat into your current coding session.

List your CoChat conversations, most-recently-updated first. Pass a `query` to search by title. Returns id + title + updated_at; use chats_get to read a conversation's messages.

Free, read-only: the paper's registry identity verdict, retraction status, and the audit of its own reference list (does each cited work exist and is it registered?). `paper` is a DOI, an arXiv id, a doi.org/arxiv.org URL, or a CoChat Library paper id. Never creates, verifies or dispatches anything — plan='refused' when CoChat doesn't already hold a publicly-identified copy of this paper (call verify_paper first to add and verify it); plan='pending' while its citation graph has not been fetched yet by other means. This never spends a paper verification.

Read the paper's own claims and each one's reference-support verdicts. `paper` is a DOI, an arXiv id, a doi.org/arxiv.org URL, or a CoChat Library paper id. If nothing has been checked yet this starts verify_paper's one metered run and answers plan='started'; call it again once the plan reads 'ready'.

Fetch the full text of a paper by the id (DOI / arXiv id / ERIC id / URL) from a search result. Reads CoChat's own indexed full text first when we already hold it; otherwise resolves a free open-access copy (arXiv, Unpaywall, Semantic Scholar, CORE, ERIC) and returns the extracted body — including on a paper's first fetch, which also adds it to CoChat's shared index. Falls back to the abstract when no open-access full text exists anywhere, or when the held text is licensed for analysis only. Never fabricate a body — trust the returned status.

Add a paper to the caller's own CoChat Library by DOI, arXiv id, or doi.org URL. Runs registry verification and global dedupe; lands in the Library's Inbox (or wherever an existing row already sits). Use verify_paper afterwards to check it.

Search the papers already saved in the caller's own CoChat Library, across every research question and project — not the literature at large (use `search` for that). Omit `query` to list the most recent. May refresh stale question-stance labels in the background.

List the plans you've shared, most-recently-updated first, each with its open-comment count so you can see which have pending feedback. Pass `project` to list only that project's plans.

Pull a shared plan and its review feedback. Returns the current markdown plus `open_comments` — unresolved comments teammates left on the document. Each comment carries `on_text` (the exact passage it's attached to), `context_before`/`context_after`, the author, the comment, and any replies. Treat every open comment as feedback to address, then call plans_update with the revised plan.

Share an implementation plan with your team for review. Creates a collaborative CoChat document (markdown) and returns a review_url your teammates and other AI models can open to read it and leave inline comments. Pass the FULL plan as markdown in `content`. Set `project` to the codebase/repo name (derive it from your git remote) so plans group together. Returns a `plan_id` — save it and call plans_pull later to collect feedback.