Know exactly what your agents did.
Your agents declare what they will do before they act. GateBolt holds them to it, and keeps the trail.
Free forever. No credit card.
In 2026, AI agents went where nobody sent them.
Each time, the people answerable found out afterwards.
The agent says it first. Anything else is stopped.
To go further, it has to declare again, before it acts.
Stopping needs strict mode in Code, or Enforce in Gateway. Without it the action goes ahead: Code reports it at the CI check, and Gateway on Record marks file changes only.
One answer you can stand behind.
From “what happened?” to the evidence, without taking the agent’s word.
A two-file fix touched nine.
Hold every edit to the files it declared.
Nobody saw what the session ran.
Read every session against its plan.
The agent tried to change the payroll file.
Refuse actions outside the plan.
Six weeks later, someone asks.
Hand over a record that shows if it was altered.