The Librarian
The Librarian is an AI agent that your workspace has chosen to answer questions from its wiki. When one is set up, the Wiki home turns into a chat: you ask a question in plain language and it searches the wiki, reads the pages that matter and answers with numbered sources that link back to them.
It answers from wiki pages and from files uploaded to the wiki, using only what you can open. For pages it reads only the published version. If the wiki doesn’t cover something, it says so briefly rather than guessing.
Asking a question
Click Wiki in the left sidebar. Where you would normally see the wiki landing, you’ll see Ask anything about your workspace’s name, with the Librarian’s name and photo above a box that says Ask the Librarian a question…. Under the box, Suggested questions give you a place to start, such as “What are our key policies?” or “Where do I find onboarding material?”.
Type your question and press Enter, or click Send question. Shift+Enter starts a new line. A question can be up to 8,000 characters.
While it works, a short status line tells you where it is: Getting started…, Searching the wiki…, Reading pages…, then Writing… as the answer streams in. If several answers are already being written in your workspace, you may see Waiting for a free slot… first; your question is queued and starts as soon as one finishes. You can ask again once it finishes; until then the box says The Librarian is answering.
You can keep going in the same conversation. The Librarian remembers what was said earlier in it, so a follow-up like “and what about contractors?” works. If you later lose access to a page that an earlier answer drew on, that part of the conversation is left out of what the Librarian remembers, so it can’t be used to bring the page’s content back.
Reading an answer
An answer carries numbered markers such as [1] and [2]. Click one, or one of the entries in the Sources row under the answer, to open that page. Sources are numbered in the order the answer first mentions them, and the Sources row lists them in the same order.
A source reads Source no longer available if the page has since been deleted or you’ve lost access to it. Nothing about the page, not even its title, is shown to you in that case.
Under an answer:
- Copy puts the answer text on your clipboard.
- Retry appears if the answer failed (The Librarian couldn’t finish this answer.) and tries it again.
Answers are plain text with simple formatting. They never include pictures, and any link in an answer is shown as text rather than something you can click, so nothing in a page can turn an answer into a way to send you somewhere else. The only links are the numbered sources.
What it can see
The Librarian answers for you: it can read only what you could open yourself. A page that is Restricted and not shared with you is never searched, read or quoted for you, even though the Librarian itself might be able to open it for someone else. See Sharing and access.
It also reads only the published version of a page. That is true for everyone, including the people who can edit it: your unpublished changes are not used, and a page that has never been published is not used until it is published. See Drafts and publishing.
Uploaded files
The Librarian can answer from the text inside uploaded files: PDFs with selectable text, Word (.docx), PowerPoint (.pptx) and Excel (.xlsx) documents, and plain text, Markdown, CSV and JSON files. Older Office formats (.doc, .ppt, .xls) can be imported but not read; save them in the newer format first. Answers cite the file as a source, and clicking the source opens the file.
It can’t read scanned documents (a PDF that is only pictures of text) or images yet, and it doesn’t read video. They can sit in the wiki, but the Librarian won’t find anything in them. Files larger than 10 MB aren’t read at all, and a very long document may be read only in part. They still import, open and download normally.
A file you import is usually ready to answer from within moments; the import dialog tells you when it is. Files uploaded before the Librarian could read files are read gradually, a few each time someone asks a question, so for a while some of them may not be found. To add many files at once, see Importing files.
Guests can ask too. Their answers draw only on the pages shared with them.
Turning files into pages
When you import files, the Librarian can also turn some of them into wiki pages instead of keeping them as files. You choose this in the import dialog, and nothing starts until you confirm. This uses the Librarian’s Daily token budget, and the work shows in its Runs tab. The Tokens used by Librarian answers today figure does not include import tokens, while the daily budget counts both. See Turning files into pages.
Your conversations
Every conversation is private to you. Other members and workspace administrators can’t read your conversations in 3mote, and they are kept separate from the Librarian’s direct messages in Chat.
- History in the page header lists your conversations, most recent first. Click one to reopen it; Load more shows older ones.
- New conversation starts a fresh one.
- Coming back to the Wiki in the same browser tab, for example after opening a source, pressing Back or visiting another part of 3mote, reopens the conversation you last had open. New conversation or deleting the conversation starts you fresh.
- To delete one, click its delete button in History and confirm Delete this conversation?. The conversation and its answers are permanently deleted and can’t be brought back.
A conversation is removed automatically once it has been idle for longer than your plan’s chat history period (Free 30 days, Team 3 years, Enterprise 10 years). See Plans and storage.
Limits
You can ask 30 questions an hour. The limit is per person and counts across all your conversations, including ones you’ve deleted. If you reach it, you’re told how many minutes to wait.
When it isn’t available
Instead of the question box, you may see one of these notes. Your earlier conversations stay readable in each case.
- The Librarian isn’t available right now. (The note uses the agent’s own name, such as Qin isn’t available right now.) Its seat is inactive, or it no longer has access to the wiki. Ask a workspace administrator to check it.
- The Librarian is paused. Its model needs a plan upgrade before it can answer.
- The Librarian has reached its usage limit for today. It can answer again tomorrow, or an administrator can raise its daily limit.
If your workspace hasn’t set up a Librarian, the Wiki shows its usual landing page.
What is sent to the AI provider
Asking the Librarian is not private from the AI provider your workspace uses. To answer, 3mote sends that provider:
- your question;
- a recap of the conversation so far, with the recent turns; and
- the text the Librarian searches and reads to answer (excerpts, passages and whole pages or files), and the names of the pages and files it lists.
The provider is the one your workspace’s administrators configured for that agent, under the workspace’s own account with them. They choose and control it, its terms apply to what it receives, and 3mote can’t promise what it does with the content.
You can always see which one. Under the question box, the Wiki page says, for example, “Questions and the wiki pages used to answer them are sent to” the provider and model. If your workspace uses its own AI endpoint, it names that endpoint’s address instead. Learn more opens this page. The model also appears under the Librarian’s name at the top of the page.
Inside 3mote, your conversations stay private: other members and workspace administrators can’t read them. See the Privacy Policy for how conversations are stored and for how long.
When the Librarian’s provider, model, address or key changes, or a different agent becomes the Librarian, the workspace’s owners and administrators (other than the person who made the change) get a notification, and the change is recorded in the agent’s history.
Setting it up
Setting up the Librarian is for people who can manage AI agents, which the Account Owner and Administrators can by default.
If no Librarian is set up and your workspace already has an AI agent, an administrator opening Wiki sees Choose a Librarian: pick one of your AI agents in the AI agent list and click Make Librarian. Everyone in the workspace can then ask it from that page. A workspace with no AI agents doesn’t see this; the Wiki page stays as it is.
To make a new agent for the job, open Settings → Users, click Add AI agent, and choose the Librarian persona, which starts the agent with instructions suited to answering from the wiki and citing its sources. In the dialog, tick Make this agent the workspace Librarian. See AI agents for the rest of the dialog.
Once an agent is the Librarian, a book icon follows its AI tag wherever the agent is named, in chat, on cards and in Settings → Users. Hover the icon to see Librarian.
You don’t need the persona. Any active AI agent can be the Librarian, and you can switch an existing one on or off from Settings → Users → Edit, using Workspace Librarian in the agent’s settings.
To be the Librarian, an agent must be active and its role must be able to view wiki pages and create folders. Of the roles an agent can hold, Manager has both by default; if you use a different role, check its permissions under Settings → Roles. An agent that doesn’t qualify is refused with a message saying so.
A workspace has one Librarian at a time. Making another agent the Librarian asks you to confirm Replace the Librarian?, and earlier conversations stay readable. Turning Workspace Librarian off returns the Wiki to its usual landing page.
The agent’s own limits apply to the Librarian, including its Daily token budget. See AI agents. When the Librarian reaches that budget, the agent’s accountable person gets one notification that day; if that person is no longer active, the workspace’s owners and administrators get it instead. The notification doesn’t say who asked or what.
Related
- Wiki — the page tree and the row menu.
- Importing files — adding documents the Librarian can answer from.
- Search and action items — finding pages yourself.
- AI agents — adding and configuring an agent.