What would you like to secure?
AI agents in production
Can you stop an AI agent, and prove what it did?
An agent that can approve a claim or release a payment can also do the wrong thing. Sigill Guard stops actions your policy forbids before they run, and every run leaves a record that can be checked independently of the system that ran the agent.
03 · How Sigill helps
Every step sealed as it happens, the run closed with a timestamp.
01 / 04
Your agentControl basis
Before the agent acts, its instructions, tools, policy and model settings are sealed. That fixes the rules it started with.
What the evidence proves
- The recorded steps are unchanged since they were sealed, and in the recorded order.
- The agent's instructions, tools and policy were sealed before the first step.
- When the run was closed, according to an independent timestamp authority.
What it does not prove
- That every step the agent took was recorded. Only what was recorded is covered.
- That the agent's inputs, or its own statements, were true.
- That the agent's decision was the right one.
04 · Go deeper
Technical depth, when you want it.
Policy and enforcement Where the guard runs, a policy example, and what happens when Sigill is unreachable.Agent evidence in the API docs Recording runs, policies, approvals and findings.Open-source SDKs Python and .NET recorders and verifiers, Apache 2.0.Verify a run bundle Check an exported run in your browser, no account needed.
Try it yourself, or run a pilot
Create an account and an API key to evaluate Sigill on your own, or talk to us about a pilot.