A narrow product, built carefully, for people with a lot to lose.

Hi — I'm Hongyi, the founder of Attestum.

I started building this after a year of watching production AI agents cause real incidents — the kind that reach regulated customers, ship the wrong tool call, or land the wrong field in a downstream system that auditors look at. Every post-mortem concluded the same way: we need better evals and faster rollback.

Meanwhile the industry kept shipping observability. Prettier traces. LLM-as-judge rubrics. Dashboards with sparklines. All of it useful, none of it enforcing. Observability is forensics — governance is what stops the incident mid-run. That's the gap Attestum closes.

The product is deliberately narrow. We are not an AI gateway. We are not an observability platform. We are not an agent framework. We are the control plane that turns your signed SLA into enforced behavior, with an audit trail your compliance team can live with.

I'm building in public. The open specification at github.com/attestum-ai/agent-sla-spec is intentionally free and framework-neutral. If it's right, it should be implementable by five different vendors by 2027. Attestum is the first. Adoption is the moat — not proprietary lock-in.

If your agents are going to production and your compliance team is asking questions you don't have documented answers to yet, I'd like to talk. Request early access or email me directly at [email protected].

— Hongyi