Buff without a repository: where to start

This page is for an employee of a department that has no repository and needs none: support, sales, legal, HR, finance, marketing, a manager, a product manager. What you get:

• answers to questions about the department’s documents — policies, contracts, spreadsheets, wiki pages — with a link to the document and section; if two documents disagree, both are named;
• documents in which the AI agent has reconciled several sources and named the discrepancies — deep research;
• the department’s knowledge base, where a rule is written down once and taken into account in every answer.

The AI agent only reads: documents are not changed, emails are not sent, tickets are not created. The track has six steps; one of them, connecting the documents, is done by your IT specialist.

1. Account and organisation

Register and sign in — Registration and sign-in. If the company already has an organisation in Buff, register and ask the owner to invite you: they will give you a role and a scope — usually member (Contributor in the interface) on your space (see Managing members). If there is no organisation yet, create one — Organisations.

2. Space

A space is a workplace without code: “Ask”, deep research, a knowledge base and connections to documents. It is created from just a name — here is how.

A space can be created by the organisation owner or a member who has been given the organisation-wide permission to create projects and spaces (in the interface this permission is called “Add repositories and sync the project” — yes, spaces are created with it too). If you do not have this permission, ask the owner to create the space or to grant you the permission.

A space is a permission scope, like a project: it makes sense to have one space per department. Then HR documents are read only by members with permission on the HR space, and support does not see them until it is given access.

3. The department’s documents

The documents stay where they are: in Notion, Confluence, on Google Drive, on a server or in a network folder, in Figma. Nothing is copied in advance — the AI agent reads only what was needed for the answer.

What your IT specialist does once:

1. installs the connector — a small program in the company network; one connector serves all departments;
2. sets up the right connection from the catalogue — Notion, Confluence, Google Drive, files on a server and network folders, Figma, a tracker, the internet — or your own tool for a system that is not in the catalogue;
3. names your space in the connection as the place where the work goes;
4. selects the sections and folders that may be read.

What to hand over to them, point by point — What to ask your administrator for.

What you can do yourself:

• on Google Drive — share the folder with the account whose address IT gives you, with the “Commenter” permission;
• in Notion — add the integration to the section: on the section’s root page, menu “…” → “Connections” → pick the integration; everything nested opens along with it.

What is read: Word, Excel and PowerPoint documents (new and old formats), PDFs with a text layer, text files; Google documents, spreadsheets (a sheet and a range) and presentations; Notion pages and databases; Confluence pages; Figma design properties. Not read are scans without a text layer and images: there is no text recognition.

4. The first question

Open the space → “Ask” → a new question. Name the document, section or period if you know it, and say what form the answer should take. Examples:

• “What first-response time does the contract promise a client on the ‘Business’ plan, and does it match the FAQ?”
• “How do I arrange a holiday, and how many days in advance do I submit the request?”
• “Which supplier payments fall in October, and where is there a prepayment?”

The answer comes with a link to the document and section; discrepancies between documents are named. Attach a file that is not in the sources to the question — up to 10 MB. If the answer is “not found”, the How to ask a good question page explains what that means.

5. Research

When you need not an answer but a document — to bring several sources together, find all the discrepancies, prepare something that will be forwarded — order a deep research. Two roles of the AI agent: the analyst writes, the checker verifies every claim against the sources. The result is a document with links and the discrepancies named, exportable to PDF and a text file. How this differs from a question and when to choose which — Question or research?.

6. The first knowledge base entry

The knowledge base holds the department’s rules, facts and pitfalls; the AI agent sees them in every question and research. Create one or two entries with the “space” scope:

Writing to the base is open to a member with the “Create, edit, archive and restore knowledge” permission — by default the space editor has it (Maintainer in the interface); all members of the space can read. The AI agent also adds facts and pitfalls itself, with a mark — AI entries.

Next

• Recipes by role — what to connect and what to ask for support, sales, legal, HR, finance, marketing, a manager and a product manager.
• Ask right in Notion, a Google Doc or a tracker — without opening the app.
• What a question costs — what makes up the price and where to see it.
• Security: what leaves and what does not — for the owner of the documents.