Connect a source
Put a GitHub, Google Drive or ClickUp connection in place so Brain is allowed to read a source, and know exactly what it will read from it.
Brain reads GitHub, Google Drive and ClickUp, and it holds no credentials of its own for any of them. It reads through a connection somebody in your organization installed and made available to your product. Until that connection exists, there is nothing for Brain to read.
Make a connection available to your product
Connections live on the organization, not inside Brain.
- An organization owner or admin installs the connection once. See set up an integration.
- The same owner or admin attaches it to the products allowed to use it.
- If you are not an owner or admin, request the connection you need and wait for a decision. See request access to an integration.
A public GitHub repository needs no connection at all. Point a daemon at the owner, repository and branch, and Brain reads it anonymously. Anything private needs a connection.
Connecting a source only grants access. It moves nothing into memory. What is read, from which repository, folder or space, and how often, is the daemon — see ingest a source on a schedule.
When Brain lists what you can pick from — GitHub accounts and repositories, Google Drive accounts and picked files, ClickUp workspaces, spaces, lists and chat channels — it is listing what those connections already reach. Brain cannot widen that. A repository or folder missing from the list has to be added to the connection on the organization side.
What each source gives Brain
| Source | What is connected | What access that requires | What Brain reads | What Brain does not read |
|---|---|---|---|---|
| GitHub | A GitHub App installation on the accounts or organizations whose repositories you want read. Brain asks that installation for repositories and branches, and reads content with a short-lived token minted per run — it stores no GitHub credential itself. | Read access to the repositories you intend to ingest, including their issues and pull requests. Brain never writes to GitHub. | Markdown, issues, pull requests. | Code files other than Markdown. Comments, review threads and review state on issues and pull requests. Commits, diffs, releases, wikis, projects, discussions and actions. Anything outside the one branch the daemon selected. |
| Google Drive | A Google account, together with the specific files and folders picked for it. The picked set is the entry point: Brain reads those items and nothing else in the Drive. | Read access to the picked files and folders. Brain never writes to Drive. Shared drives are supported; items in the Trash are skipped. | Google Docs, Sheets and Slides; PDF files; plain text files. | Any file type outside that list — images, audio, video, archives, Google Forms, Google Drawings, and binary office formats. A PDF with no extractable text, such as a scan without an OCR layer, is skipped rather than ingested empty. Comments and suggestions. Revision history. |
| ClickUp | A ClickUp workspace. Brain lists that workspace’s spaces, lists and chat channels so a daemon can be scoped to them. | Read access to the tasks, docs and chat of the workspace you intend to ingest. Brain never writes to ClickUp. | Tasks, comments, docs, chat. | Archived spaces and archived lists. Task attachments, checklists, time tracking, dependencies and activity history. Whiteboards, dashboards, goals and forms. Comments on docs. Chat in channels you did not select; chat attachments that are not plain text are recorded as a link and a title, not as content. |
Each category below is an independent switch on the daemon, so a source can be connected in full and ingested in part.
GitHub categories
- Markdown — every
*.mdfile, plus any file namedREADMEwhatever suffix it carries, on the selected branch. Include and exclude patterns narrow this further; the default is everything matching**/*.md. - Issues — all issues, open and closed, with their title, state, labels and body.
- Pull requests — all pull requests, open and closed, with their title, state and body.
One daemon reads exactly one branch of one repository.
Google Drive categories
- Documents — Google Docs exported as Markdown, Google Sheets exported as CSV, and Google Slides exported as plain text.
- PDF files — read as their extracted text.
- Plain text files —
text/plain,text/markdownandtext/csv.
Folders are followed into their contents when recursion is on, which is the default. With recursion off, a picked folder contributes nothing and only picked files are read.
ClickUp categories
- Tasks — name, status, assignees, non-empty custom fields and the task body.
- Comments — the comments on each ingested task, with author and date.
- Docs — each doc in scope, page by page.
- Chat — messages in the channels you selected, with their replies, the links they contain, and the text of plain-text attachments. Chat is off by default.
A daemon is scoped to one workspace, optionally narrowed to a single space, or to a single list inside a space. With no space selected, every non-archived list in every non-archived space is in scope. Chat channels are selected individually.
Chat is the one category with a history bound. When a daemon first syncs a channel it loads the last 7, 30 or 90 days, or all history, according to the initial load window you chose; 30 days is the default for a new daemon. Every later run reads only messages newer than the newest one already ingested from that channel. Narrowing or widening the window afterwards does not re-read what has already been passed.
Trust of what is ingested
Everything a daemon ingests is stamped with a source trust level, and the level follows the kind of source it came from.
| Ingested from | Stamped as |
|---|---|
| Repository Markdown, Drive files, ClickUp docs | official |
| Issues, pull requests, tasks, task comments | team |
| ClickUp chat | informal |
A daemon can override this for everything it ingests, including stamping it policy — see ingest a source on a schedule. What the levels do to search results is described in what a memory is.
When a connection is withdrawn
A connection that is detached from the product, or uninstalled, stops working for every daemon that used it. Those daemons keep their configuration, but their runs start failing on access errors. Memory that was already ingested stays where it is.
Next steps
- Ingest a source on a schedule: turn a connection into memory that keeps itself in step with the source.
- Request access to an integration: ask an owner or admin to attach a connection your product does not have.
- Set up an integration: install a connection on the organization and attach it to products.
- What a memory is: what the source trust levels do to a search result.
- Look up a Brain tool: the tools your agents call to work with what the product remembers.