About Atested
We built it because we needed it built
Atested is a governance layer for AI agents and applications. Today you carry the responsibility for what your AI does but have no way to know, much less prevent, what it's actually doing. With Atested, whole categories of errors are prevented. When something does happen, you have certainty about what happened and why.
What we all know
The downside is equally impressive
AI can do real work. But what agents say they do and what they actually do can diverge. The more capable the agent, the larger the blast radius of a mistake. Atested is built to tame the downside. Prevention is better than painstaking forensics.
Why we built it
We use AI agents every day. We had the same problem you do.
We thought an MCP server was the way to go
We were wrong. If an agent can go around something it eventually will find a reason to and they went around it. API connection is structural and our eager AI buddies don't even know its there so they don't even try to avoid it. Problem solved.
The policy is obvious once we thought about it
And being obvious means no one has to recreate the wheel.
Cryptographic chains are cool
And they are much better than logs. Logs can be edited, deleted and questioned, none of which is true about our chain. It is Ed25519 signed, hash-chained records and if you know you know. There is no arguing about them. They are cool.
Atested is good but everything has limits
No one, not us or anyone knows what is true before it happens. What we can know is if the process was flawed. That's what Atested does it prevents flaws in reasoning. We insist the AI explain its reasons and if they are flawed we stop the AI. That's the most anyone can do.
What we believe
Deterministic methods are best
Deterministic methods are certain. Where there is uncertainty there is judgment which implies a choice which means at least one choice is incorrect. We reduce uncertainty by exhaustively using deterministic methods. When that is no longer possible we carefully employ judgment.
Fit over business
Our case document is a living document that informs you of your Atested usage. We don't game the system to constantly upsell you. We hate that.
Don't police AI thinking
Atested doesn't require you to change how you prompt or how your agent is configured. It watches and intervenes when necessary, in a way that improves AI operation. And when the worst happens, Atested is there with the record.
One more thing
The product isn't the dashboard
The product is the proxy. The HTTP connection that sits between your agent and its model, evaluating every model-proposed action before your agent receives it. The dashboard shows you what the proxy did. You will use the dashboard but that's just the window. The real work goes on in the background, but you knew that.
Trust and integrity
We don't ask you to trust us
The chain is on your machine, signed with your key, and anyone can audit it independently. Telemetry is transparent — you see exactly what leaves your machine. Every proactive channel has an opt-out.
How we build
Atested is built with AI
Not assisted by AI. Built. Every file in the repository has AI involvement. The way through the failure modes is discipline — bounded tasks, clear success criteria, human verification before anything ships.
Start attesting your AI operations
Stop worrying and start knowing in less than five minutes.