> ## Documentation Index
> Fetch the complete documentation index at: https://supermemory.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# The Permissions Graph
> What Company Brain remembers, who it's visible to, and how tool access is scoped
Company Brain isn't split into "a shared brain" and "a private brain." It's a graph: memory is written to the narrowest room a conversation happened in, and what a given conversation can *read* depends on where it's happening and who's asking. Nothing here is silent — every install, channel read, and temporary access grant requires an explicit accept from a real person.
## Three memories, not two
One per person. Built from your DMs with the bot and what it learns about you over time. Only visible from your own DM.
One per private channel. Scoped to that room — visible to anyone in it, to no one outside it.
One per organization. Anything durable from a public channel lands here. The whole org can draw on it.
A message writes to exactly one of these — whichever room it happened in.
## What a conversation can read
Writing is narrow; reading is broader, and it widens the more private the room is:
| Asking from | Can read |
| ----------------- | ----------------------------------------------------------------------------------------- |
| A public channel | Public channel memory |
| A private channel | That channel's memory + public channel memory |
| A DM with the bot | Your employee memory + public channel memory + every private channel memory you belong to |
```mermaid theme={null}
flowchart LR
Pub["Public channel memory
(the whole org)"]
Priv["Private channel memory
(that room's members)"]
Emp["Employee memory
(you, in DM)"]
Priv -.reads.-> Pub
Emp -.reads.-> Pub
Emp -.reads.-> Priv
```
A DM is the widest seat in the room precisely because it's the most private one — the bot answers you there with everything *you* could see, stitched together. A public channel is the opposite: the whole org can read it, so it only ever draws on what the whole org is allowed to know.
If you're not in a private channel, its memory doesn't exist for you — not even by inference in a DM. The bot only ever reads with the asker's own access, so it can't surface something you couldn't otherwise see.
**Example:** you DM the bot asking "what did we decide about the Acme deal?" It can draw on the public `#sales` channel, the private `#acme-deal` channel if you're in it, and anything it's learned about you directly — and it'll cite which one the answer came from. Ask the same question in `#general`, a public channel, and it can only answer from what `#general` and other public channels already know — the private `#acme-deal` context simply isn't in scope there.
## Tool access follows you, not the connection
Tools like GitHub and Linear can be connected two ways — **Organization (shared)**, set up once by an admin as a fallback the whole team can read from, or **Personal (yours)**, your own connection for your own reads and actions. Both show up on the same connections page; it's one tool catalog, connected at two possible scopes.
Whichever scope answered, the result is still bounded by what *you* could already see or do in that tool yourself — Company Brain never gets a standing key to "everything Linear knows." If you're not on a private Linear team, the bot can't surface those issues to you either, even through the org-shared connection.
| | Reads | Writes |
| ------------ | ---------------------------------------------------------------- | ---------------------------------------------------------------------- |
| **Behavior** | Try your personal connection first, then fall back to org-shared | Always run under your own connection |
| **Why** | Gives you the fullest access you're entitled to | Attributes the action to a real person, never a shared service account |
Admins can also act through the org-shared connection directly, for the cases where that's the point.
## Leasing: borrowing access for one request
Sometimes a request needs a tool neither you nor the org has connected — but a teammate has it connected personally. Rather than failing, Company Brain can ask that teammate directly: it posts a card in Slack asking them to approve or deny lending access for that one request.
* Nothing is granted silently — a real person has to accept the card.
* Access is short-lived and scoped to the single request that triggered it, not standing access to your account.
* The teammate can say no, and the request simply doesn't go through.
Leasing is a fallback of last resort — it only comes up when nobody's connected the tool at the org level yet. See [Connectors](/docs/company-brain/connectors) to close that gap for good.
## API keys inherit the same graph
A scoped or agent API key can only reach what its owner could already reach by asking directly. A member can't mint a key that reads another member's employee memory or a private channel they're not in — the graph above applies identically whether a person is asking or a key is.
Set up the data and tool connections this page describes.
How scheduled runs and unprompted replies respect the same graph.