Built by Costa Xanthos in 2026.

AI agents are about to be the dominant traffic shape on the internet. Today’s control plane — rate limiters keyed by IP, identity providers built around one-human-one-token, observability tools that only describe what happened — was built for a world where requests originate from people clicking buttons.

Agents fan out. One goal becomes thousands of sub-requests. They share keys, repeat themselves, and casually take down their own databases. Your infrastructure interprets this as a DDoS attack and either crashes or rate-limits everything indiscriminately.

Vigil is the agent-aware data plane. It sits between the agents and the systems they touch and shapes traffic the way agent traffic actually behaves. Per-agent identity. Smart rate limiting that knows which agent is which. Fan-out coalescing for the redundant queries. Blast-radius policies enforced at the proxy, not in the agent’s prompt where it can be jailbroken out of. Signed audit trail for everything that flows through.

The bet is structural: foundation labs can’t build this. They live in the LLM call path, between you and the language model. Vigil lives in the application call path, in your VPC, in front of your Postgres. That’s the part of the stack the labs can’t see.