Skip to content
Log in

How Brain works

Brain is the memory of one product: what its agents write down, what the sources you connect put into it, and the search that answers questions about all of it.

Brain is memory that belongs to one product. It holds what the product and the people around it know, keeps that knowledge searchable, and fills itself from the sources you connect to it — the agents working in your product write into the same memory.

Everything Brain holds belongs to that one product, and two products never see each other’s memories. This page explains the path from a product that has never used Brain to one whose memory is filling itself and answering questions; the mechanics of calling Brain from your own agents are not here — the platform publishes and describes the tools each product can use. See look up a Brain tool.

From nothing to an answer

Six steps, and the last one is optional. You need a product on the platform and access to it. Step 2 additionally needs somebody with organization owner or admin rights, which may not be you.

  1. Open Brain in your product. There is nothing to install: Brain is available in every product on the platform, and its search comes up empty at this point because the product has no memories yet. See what a memory is.
  2. Connect a source at the organization level. Brain reads GitHub, Google Drive and ClickUp, and holds no credentials of its own for any of them — it reads through a connection an owner or admin installs on your organization and attaches to the products allowed to use it. If you are not one, you can request the connection you need and wait for an admin to approve it. See connect a source.
  3. Set up a daemon so that source is ingested. A daemon is one source, ingested on a schedule: you pick the source type, narrow it — a repository and a branch, a connected Drive account, a ClickUp workspace — and switch on the categories you want taken from it. Two settings apply to every daemon: the source trust level its memories are stamped with, and the load mode, which trades speed against cost on large volumes. See ingest a source on a schedule.
  4. Let the first ingestion run and check that it completed. A new daemon starts enabled and is picked up by the hourly ingestion cycle, and you can start a run immediately instead of waiting. The first run is the expensive one — nothing has been read before, so every item in scope is fetched, and a large repository can take hours; later runs skip items whose content has not changed. See ingest a source on a schedule.
  5. Search your memories to see what landed. Ask a question in ordinary language: search is not keyword matching, so a memory phrased differently from your query still comes back. See add, search and correct memories.
  6. Optionally, add a memory of your own. Ingestion is not the only way in — agents working in your product write into the same memory, choosing the text, whether the memory is shared with the product or private to that agent, and how much authority the statement carries. See add, search and correct memories.

Connecting a source only grants access; it moves nothing into memory. A public GitHub repository is the one case that needs no connection at all.

Writing is done by agents, not by people. What people contribute is written through the agents they work with, or through the documents they put into a connected source. A single write usually produces several memories: the text as submitted, plus the individual statements Brain read out of it.

On a product this new, search returns one group of results — the direct ones. The other two groups are produced by cognition, which consolidates what Brain holds in the background over time, so they fill in later. That is expected, not a fault.

What you work with

  • Memory — one statement Brain holds for your product, searchable by meaning rather than by keyword. See what a memory is.
  • Scope — whether a memory is shared with the whole product or private to the agent that wrote it. See what a memory is.
  • Source authority — the trust level a memory is stamped with, which says how much weight the statement carries. See what a memory is.
  • Connector — the GitHub, Google Drive or ClickUp connection that lets Brain read a source. It lives on your organization, not inside Brain. See connect a source.
  • Daemon — one source, ingested on a schedule, with its own runs and its own record of what each run took and refused. See ingest a source on a schedule.
  • Cognition — the background work that consolidates what Brain already holds. See how memory maintains itself.
  • Knowledge graph — the entities and relations built out of your memories, searchable in their own right. See search the knowledge graph.

Next steps

Was this page helpful?
Esc

Start typing to search the docs.

↑↓navigate↵select