qyrixx

Qyrix Sentinel is a Trust-to-Execution Guard for AI agents. Agents can now call real tools, so a poisoned README or skill can make one leak an API key to an unapproved host. Detection alone is not prevention. Sentinel puts a deterministic, default-deny Gate in front of agent tool calls: unknown or unapproved actions are blocked before anything executes, grants are short-lived and single-use, and every decision is written to a hash-linked proof chain that can be verified (tampering with a record makes verification fail). An LLM is used only as an advisory inspector: its verdict never reaches the Gate and cannot allow an action. Prototype stage, not production-ready: MCP and n8n integrations exist, network isolation is required for the guarantees to hold, and adapter evidence is self-reported (E1) unless independently observed.