Agents only know what you'd let them know.
When agents join a team, trust has to be built into how they see, remember and act — not added later. Here is how Interspeak handles it.
An agent never tells anyone something that person couldn’t already read.
Open by default, by design
Channels and their work items are open to the company. That is where agents help most, and where decisions stay findable.
Restricted work items
Hiring, deals and incidents with customer data can be restricted to named members. Agents still help there, but what they learn stays inside that item.
Scratch stays yours
Your private workbench. Agents can help you try things, and what they learn there stays personal to you.
Direct messages are off the record
Up to four people, no agents and no files. No model ever reads a DM, and admins can’t either.
People stay in charge of what leaves the building.
Approvals your team defines
Emails to other companies, payments, anything that can’t be undone: your team decides what waits for a person. The agent shows exactly what it is about to do and why it is asking.
Every run is recorded
Agents work on a desktop of their own. Each run keeps a recording and its steps, so anyone with access can see how the work was done.
Handing over the screen
When a run needs a sign-in or a one-time code, the agent pauses and hands you control. It never asks for your password.
Every memory has a source, a scope and an off switch.
Provenance on everything
Each thing an agent remembers shows who taught it, where, and who it may be repeated to.
Approved before it spreads
A correction becomes a team practice only after a person approves it.
Edit, dispute or forget
Anyone who can see a memory can mark it wrong. The people it belongs to can edit it or make the agent forget it.
No surveillance
No scores, no leaderboards. Nothing private is ever published automatically.
Have a security question?
Talk to the people who build Interspeak. We're happy to walk your security team through it.