PRODUCT · RUNTIME CONTROL
CONTROLS WHAT YOUR AI DOES.
Let your AI agents act on their own. Stop the one action that should never happen. Every action an agent proposes is checked against rules written in plain English before it runs, and returns ALLOW, HOLD, or BLOCK. The model still decides. The action does not run until it has been checked.
What it does
- Stops tool chaining: blocked once, blocked again on the workaround
- Catches agent drift: the goal it was given, not the one it wandered to
- Zero-day agent behavior stopped the day it appears
- No new rule, no signature, no patch to do it
Rules in plain English, written by whoever owns the risk, with no code to change one. HOLD is an option you switch on for the actions that need a person, with the approver and the timeout declared in advance; everything else runs at full speed. Every decision is logged, who, what, when, why, on an append-only record. Model and runtime agnostic: the first published runs are on Hermes, the autonomous agent from Nous Research, and the same control reaches hosted assistants, open-weight models on your own hardware and custom agents.
Every AI system we can govern, and the ones we cannot yet
The scenario
This summer a vendor updated the hosted model behind our demo agent to chain tools more persistently, and the agent tried things nobody had seen, more than 140 attempts in a single run. Every new attempt was stopped the day it appeared, and nothing on our side changed. In the ungoverned lane of the same run, every one of them completed.
The July run, recorded, with the two actions we had no policy for
One policy, as you would write it
Take an agent a security team would build: an AI vendor risk analyst that runs a full vendor review end to end. It reads SOC 2 reports and questionnaires, extracts the controls, assigns risk ratings, opens remediation tickets and sends approval recommendations. Its policy, as written:
Policy
ALLOW: read SOC 2 reports, bridge letters and vendor questionnaires; create draft risk assessments.
HOLD for the named approver: approve a high-risk vendor; accept a compensating control; close a remediation finding.
BLOCK: modify a security policy; change a risk rating above threshold; delete audit evidence. No approval path. Log every attempt.
Why HOLD is the third outcome, and where it belongs
The ask
A 30-day paid proof of concept on your agents and your action set, in your environment, not ours. Then one flat annual price for the organization, with no per-agent count. Tests published, misses included, before anyone signs.
Book 30 minutes and bring the agent you cannot let run on its own yet
Proof