Blog

How Space Otter handles permissions

Your team’s agents can only share what they learn if the right people see the right facts. Here’s how access works.

Ehsan Patel · 8 November 2026

Space Otter is still in development. This describes how it’s designed to work.

What happens when an agent asks

Every time an agent asks Space Otter for something, the same four steps run.

  1. 1Identify the caller. Find the person behind the agent, and the groups they’re in.
  2. 2Filter by group. Remove every fact the caller can’t see before any search runs.
  3. 3Search what’s left. Look only through the facts the caller is allowed to see.
  4. 4Log and return. Record each fact returned, per user, and send it back with its source.

Three rules for every fact

  • Grouped. Each fact belongs to a person or to a group you set up in Space Otter. Group membership decides who can see it.
  • Combined. A fact drawn from several groups is visible only to someone in all of them.
  • Current. Remove someone from a group and their agents can no longer get its facts.

Sharing happens in your chat

Your tool asks you directly, so nobody leaves their work to approve anything. Every approval goes in the history. Approval is never a tool the model can call.

  • Sharing with a group you’re in: you approve it in chat.
  • Moving a fact to a wider group: a second person who manages its current group approves it.
  • An agent with no person present: it waits for the agent’s owner.

What we test

Each test is a way shared memory could show a fact to the wrong person.

  • A restricted user searching for a fact they can’t see gets nothing, including by direct ID.
  • A fact drawn from two groups is invisible to someone in only one of them.
  • A user removed from a group loses access to its facts.

What we keep, and what we don’t

We keep facts, each with its agent, person, source and date. We also keep links to documents, an audit log of every read that you can export, and old versions of a replaced fact, with an end date.

We don’t copy or store your documents. We don’t keep a fact after you delete it, including in search indexes and summaries. We don’t train models on your data.

Need more for a security review?

We can go through your requirements and share how we handle hosting, compliance and your data.

Agents learn. Teams remember.

Get started

Keep reading