Ask questions
Ask your connected assistant about your company, shape the answer you get, and know what to do when Brain finds nothing.
You don't ask Brain directly. You ask your own AI assistant, and the assistant asks GuruSup Brain. Brain returns evidence from your organization's memory, and the assistant writes the answer from it. Brain does not write answers.
Any member of the organization can do this once their assistant is connected. If yours isn't, start with the Quickstart or the MCP setup page for your assistant.
Before you start
| You need | Why |
|---|---|
| An assistant connected to Brain | Brain has no chat window of its own. |
| A built memory | Until an admin builds it, Brain has nothing to answer from. The preview on pages such as Sources shows only before the build starts. |
If the memory is not ready, your assistant is told that Brain isn't ready to answer company questions yet, and that an organization admin builds it from Overview once a source is connected. This applies while the memory is being built for the first time and while it is being rebuilt. If you are the admin, your assistant may instead offer a short interview about your company. Its answers help the first build.
If you are a member, ask an admin to build it. In the app, sections that need a built memory show "Available once your memory is built" and, for members, "An admin in your organization needs to build it first." Admins see a Get started button there.
Make sure the assistant checks Brain first
When you connect an assistant with Copy prompt on the Brain MCP page, the prompt also asks the assistant to save this instruction in its long-term instructions:
For anything about this company, use the GuruSup Brain tools before answering. Call init first, then ask_company_memory. Answer only from what those tools return, and if they find nothing, say so instead of guessing.
That is why you can ask a plain question and the assistant reaches for Brain on its own. If you connected the assistant by hand, the Brain MCP page also lists where each assistant keeps its instructions, so you can add the same text yourself. If an assistant answers a company question without consulting Brain, tell it to check GuruSup Brain and ask again.
What to ask
Ask about things your organization has written down or discussed in the sources you connected. Questions that pull from more than one place work well, because Brain keeps what it learned from Slack, Notion, HubSpot, webhooks and uploaded files in one memory.
- "What did we decide about the enterprise discount, and what does the customer in HubSpot currently have?"
- "Who owns the renewal for Acme, and what was said in Slack about their last call?"
- "What changed in our onboarding process since the Notion doc was written?"
A weak question and a better one
| Weak | Better |
|---|---|
| "What's the pricing?" | "What is our current price for the Pro plan for EU customers, and did anything change in the Notion pricing page or in Slack since June?" |
| "Tell me about Acme." | "Summarize where the Acme renewal stands: owner, last call, open risks. Use what's in HubSpot and Slack." |
The better versions name the customer, product or time frame, say which sources matter, and say what to do with the result. The assistant builds the question it sends to Brain from what you write, so the more it has, the closer the evidence.
If you ask several things at once, check that every part got an answer. The assistant is told to cover each part the evidence supports, and to say which parts it could not support.
Ask for the shape you need
The assistant decides how to write the answer. Tell it in plain words what you want.
| You want | Say something like |
|---|---|
| A short answer | "Give me just the answer, in two sentences." |
| A brief | "Write a one-page brief on this, with what we know and what we don't." |
| Help with a decision | "Help me decide. Lay out the options and what the evidence says about each." |
| The raw context | "Show me the excerpts Brain found, not a summary." |
| Extra care | "This is high stakes. Only say what the evidence supports and flag anything uncertain." |
These are requests to the assistant, not switches inside Brain. The assistant may pass your wish on when it asks Brain, and Brain records it with the question. Brain does not change which pages it returns because of it, so the wording affects how the answer is written, not what evidence is found.
What happens behind the answer
- The assistant asks Brain your question.
- Brain returns excerpts from pages of its memory, the facts behind them, and the title and reference of each page. By default it returns up to three pages.
- The assistant writes the answer from that evidence, and is told to cite what Brain returned.
One question can take several tool calls. If the first result is thin, the assistant may search, then fetch the full text of the pages that look best. Admins see these calls grouped as one entry, with the pages Brain used, under Queries. See Queries.
Brain only returns what you are allowed to see. See Sources and citations to learn what a citation points to and what it doesn't.
When Brain finds nothing
If nothing in Brain's memory supports an answer, the assistant is told to say it could not find anything, not to guess. It is also told that not seeing a fact in the excerpts doesn't prove Brain doesn't have it. So try rewording before you give up:
- Use the names people actually use for the customer, project or product.
- Narrow the question, or widen it if it was very specific.
- Check with an admin that the source holding the answer is connected and finished learning. Brain can only answer from what it has read. See Follow ingestion and How Brain works.
If some parts of your question are answered and others aren't, the assistant is told to say which is which.
An error is not "nothing found"
If Brain is unavailable or your access is refused, the assistant is told not to present that as an empty result. It should explain the problem plainly instead. If you get "I couldn't find anything" together with hints of a connection or permission problem, ask the assistant what went wrong, then check the MCP troubleshooting page.
Ask a colleague
When Brain lacks part of the answer, the assistant may offer to ask a person. Nothing is sent until the plan is clear and you agree.
The assistant previews the request
It asks Brain who could answer, without sending anything. Brain picks from the people who are set up as contacts. Admins set them up in Team, and members can set up their own under Account. It never picks someone just because they are named in a Slack thread or a document.
You see who would be asked
The assistant names the person and the question. If you named someone yourself, Brain checks that this person is a set-up contact. If not, the assistant says so. If two contacts match a name, you are asked to use the full name.
You confirm
The assistant sends the request only after you confirm. The person is reached on a contact method they set up, and only on the channels the assistant allowed.
You carry on
The request runs in the background and can take several minutes. Follow it in My requests. When the person answers, the answer becomes part of Brain's memory.
Good to know:
- Asking to be contacted yourself. If you say "contact me so I can add this", the assistant checks that you are set up as a contact and sends the request to you without asking again. If you have no contact method, you see "You don't have a contact method yet, so Brain can't contact you. Add one to your profile first." Add one under Account, see Brain contacts. If Brain would pick someone else, the assistant tells you it can't contact you and does not send the request to that person.
- No contact available. If nobody is set up as a contact, the assistant tells you there is no team contact available. It does not guess. Ask an admin to add contacts in Team, or add your own under Account.
- Who can be asked. If the question rests on what Brain found in your sources, the person must be a member who is allowed to see that knowledge. External contacts are not asked in that case.
- Choice of person. Brain reads each contact's When to ask topics. If no topic matches clearly, it can still propose a contact, so check the name before you confirm.
Save something new
If you tell your assistant a lasting fact about the company, it may offer to save it to Brain. It should write to Brain only after you ask or confirm, or approve the write in your client.
| Rule | What it means |
|---|---|
| What to save | Durable company facts. Not chat, drafts, temporary preferences, guesses or things it inferred. |
| Sensitive personal data | The assistant should not save it, unless you have explicitly designated it as company memory. |
| Format | Plain text or Markdown. |
| Size | Up to 200,000 characters per item. |
| Usage | Saved knowledge counts toward your organization's usage. If usage runs out, the item waits as Paused and is learned when usage renews. |
| Afterwards | Saved knowledge can't be edited or removed from the app. Say exactly what you want saved. |
| Timing | It is not instant. Allow a few minutes before it shows up in answers. |
To check it landed, wait a few minutes and ask a question the new fact should answer. Admins can also look for the item in Follow ingestion: saved knowledge shows under the source Manual knowledge.
What members and admins see afterwards
| Members | Admins | |
|---|---|---|
| Answers in their assistant | Yes | Yes |
| Requests they started | My requests | Follow-ups for the whole organization in Ingest |
| Every question asked, pages used, confidence | No | Queries |
| Saved knowledge | Ask again to see it in answers | Also in Ingest, as Manual knowledge |
Related
- MCP tools reference: every tool the assistant can use.
- My requests: follow questions sent to colleagues.
- Brain contacts: who Brain can ask, and how.
- Sources and citations: what a citation points to.