How Sentra enforces permissions on every fact
A fact in Sentra inherits the permissions of the system it came from, checked when someone asks, not when the data was loaded. If Sentra cannot confirm access, the answer is no access.
In short
- Each fact is visible only to people who can see its source, such as the Slack channel, Drive file, Jira issue or meeting it came from.
- For Slack, Linear, GitHub, Google Chat and Jira, access is checked against the source system itself. Any error counts as no access.
- Where Sentra keeps a copy of membership, such as Teams chats and Confluence spaces, it denies access once that copy is more than two hours old.
- Agents acting for a person get exactly that person's access. Raw email bodies need a separate permission that no built-in role includes.
Why permissions have to work fact by fact
Company memory is only safe to roll out if it never shows someone what they could not already see. Permissions by tool are too coarse: a single memory can hold facts from a private HR channel and a public all-hands. So Sentra applies access to each fact, from the source that fact came from.
How access is checked
Every check fails closed. If the source system errors or Sentra's copy of membership is out of date, the answer is no access rather than a guess.
- Live checks against the source for Slack, Linear, GitHub, Google Chat and Jira, cached for up to 30 minutes
- Synced membership for Microsoft Teams, Jira, Confluence and Basecamp, denied once the sync is more than two hours old
- Drive files follow each file's own sharing
- HubSpot records follow ownership and team
- Manual notes and personal email are visible to their owner only
- An entity appears in results only if the person can see at least one fact about it
Withheld looks like absent
When a fact is withheld, it is simply not there. Sentra does not tell a person that something exists that they cannot see, so a question about supplier renewals can return every contract to finance and nothing to a sales rep, with no hint that anything was hidden.
The same rules for agents
Claude, ChatGPT, Cursor and internal agents reach Sentra through the MCP server and API, and every call is checked as the person the agent acts for. Raw email bodies need a separate email access permission that an admin adds to a custom role. No built-in role includes it, and it never widens which messages someone may see.
Meetings and email follow roles where a revoke always beats a grant. Members see meetings they attended. Admins see the organization's meetings except private meetings they did not attend.
Frequently asked questions
Does Sentra respect Slack and Google Drive permissions?
Yes. A fact from a Slack channel is visible only to people who can access that channel, checked against Slack itself, and a fact from a Drive file follows that file's sharing.
What happens if Sentra cannot check a permission?
It denies access. A source error or an out-of-date membership copy counts as no access.
Can admins see everything in Sentra?
No. Admins get no bypass on chat, project and document sources, private meetings they did not attend stay hidden, and raw email bodies need a separate permission.
Do AI agents get more access than people?
No. An agent acting for a person gets exactly that person's access, and tools the agent is not entitled to are hidden from it.