In a refinery, you never trust a signal you can't trace. Every gauge has a serial number, a calibration date, and a sign-off chain. That's not bureaucracy — it's what keeps people alive.
AI agents are being connected to enterprise workflows, supply chains, and operational systems with none of that. No standard way to verify who built an agent, what it's actually capable of, or whether it's been tampered with. Carapace is the calibration standard the agentic web is missing.
ARIA is the live registry backend. Not a dashboard you log into — an endpoint agents can query at runtime. Structured identity, capability declarations, and trust metadata, accessible by builders using the protocol.
Independent evaluator attestations are part of the Carapace roadmap. The trust model keeps the claim separated from the claimant so beta records can evolve toward enterprise-grade review without hiding provenance.
We built for the refinery floor because that's where the stakes are highest. If Carapace works when a misconfigured agent can trigger a process upset, it works everywhere.
Trust infrastructure that only works in low-stakes environments isn't trust infrastructure.
The ARIA registry is becoming the discovery layer for industrial AI tooling — the place where operators, integrators, and enterprises find tools they can actually trust to run near live processes. If your tool belongs in that conversation, there's a path in.
The A2A marketplace for industrial AI tooling. Not a curated catalog someone maintains — a trust-gated discovery network where verified tools find the environments that need them, and operators know exactly what they're letting through the fence.
Tool submissions open May 2026. Get on the list early →
We review every submission. Industrial and safety-critical tools are prioritized.