The Library
Everything your team saves lands here: market research, competitor deep-dives, documents, notes, canvases and files. Search it, preview it, download it, and read it from your own AI tools.
Updated 2026-10-13
The Library is the permanent home of what your team makes. A computer’s files are scratch; the Library is the record.
What is in it
| Kind | What it is |
|---|---|
| Market research | The report from setup or from a later research run. |
| Competitor deep-dive | A deep look at one competitor, opened from your research. |
| Document | Written deliverables such as blog posts, emails, scripts and copy. |
| Note | Short saved text, including notes saved from a connected AI app. |
| Canvas | A canvas from the old task workspace, kept read-only. Boards you make now live on the canvas page. |
| Files | Finished agent files: Excel, Word, PowerPoint, PDF, CSV, images and sites, each with its owner, version and size. |
Using it
Filter by All, Decks, Documents, Spreadsheets, Images and sites, Other files, Research, Canvases or Notes, or search by title.
- Open any item for a full preview. Documents, PDFs, slides, spreadsheets, tables, images and HTML all preview in place. A file’s preview is the one saved with it. Older files get theirs made the first time you open them.
- Download a file or export a document as a PDF.
- Delete what you no longer need.
- Files show as tiles with a picture of the file, who made it and when, so a deliverable is easy to find.
A Library with nothing in it explains why: for example, it offers to run market research if you have not yet.
How things get in
- Agents save with
save_to_library, and every deliverable they finish is archived automatically. - Research and competitor deep-dives save when they run.
- Your own AI tools save with
save_note,save_chatandrun_research. A note or chat saved for a plan task also shows on that task, under Saved from Claude (or your other AI apps).
Canvases from the old task workspace
Before tasks moved into HQ, each plan task could open a workspace with a canvas of outputs. Those canvases were copied here when the workspaces were retired. They open read-only under Canvases, as they were. Nothing runs from them any more. To act on one, copy what you need into a new chat or hand-off.
Reading it from outside
Search and read the Library from Claude, ChatGPT, Cursor or Codex with list_artifacts, search_library and get_library_item. The text comes back marked as untrusted. See MCP: Library and memory.