Rules your AI agent can't can't break

Decide what every AI agent may do, enforce it wherever the agent runs, and keep proof of what each agent did.
You're on the list.We'll email you before launch.
{{ err }}
{{ waitlistLabel }}
Trusted by Velocity University of Waterloo RBCx

It's 1 a.m. Do you know what your agent is doing?

Coding agents keep running after you log off. Nothing stops a wrong command, and afterwards no one can tell what the agent did.

Read the report · GitHub issue #93099 ↗

Has your agent ever deleted something, then said sorry?

Sorry doesn't undo it.

Real incidents, with sources. Instructions in a prompt stopped none of these.

Know a story we missed? Send it to us.
{{ storyMsg }}

Meet tego.

Write your rules once. tego applies them from the agent’s very first command.

Scan.

What the agent is given, before it runs.

Instructions
  • Load rule files
  • Detect when a rule file changes Soon
Tools
  • Approved tool and MCP server list Soon
  • Version check against the approved version Soon
Permissions
  • Granted access compared with used access Soon

Control.

What the agent can reach.

Network kernel
  • Allowed destinations
Services proxy
  • Credentials held outside the agent Soon
  • Allowed operations within a service Soon
Models proxy
  • Spending limits
  • Allowed providers Soon

Protect.

What the agent can do, at the moment it acts.

Files kernel
  • Read, write and delete, by path
Programs kernel
  • Run, by program
Tool calls SDK
  • Call, by tool and its inputs Soon
Feedback SDK
  • Reason sent back to the agent

Contain.

How far one run can spread.

Process scope kernel
  • Same limits on every process the agent starts
Privilege kernel
  • No escalation
  • No acting through more powerful programs
Persistence kernel
  • No work left behind after the run

Investigate.

What you can know about runs.

Live
  • State of every agent and subagent
  • Stall detection
Per run
  • Summary
  • Full record of actions, refusals and the rule behind each
Across runs
  • Patterns in what gets blocked Soon
  • Unused access Soon

Decide.

What happens when the rules don’t cover something.

Escalation
  • Route to the right person, where they work Soon
Learning
  • Turn an answer into a rule Soon

Three places your rules hold.

Each layer stops a different kind of wrong move, and the agent keeps working.

Tool calls. Soon

Every tool call is checked against your rules before the call runs.

Network.

Every connection is checked against the places you allowed.

Kernel.

Every file and program is checked by the operating system.

Stops the move, not the work.

Fair questions.

{{ f.a }}