Infrastructure for trustworthy AI agents
We build open-source security and orchestration tools for the MCP ecosystem. Our focus is gateway-enforced security — authentication, access control, content safety, and audit — so that AI agents can be trusted with real-world tools.
Our flagship project. A secure MCP gateway daemon for Linux that sits between AI agents and local resources.
- 22-crate Rust workspace
- Deny-wins ACLs with path canonicalization
- Information Flow Control with Bell-LaPadula verification
- In-process ML safety filters (ONNX, no GPU required)
- 138 Kani proofs, 6 TLA+ specs, OWASP ASI 10/10
- Multi-agent flows with IFC-gated mesh communication
Repository · Documentation · Why navra?
We publish peer-reviewed research on AI agent security:
- Gateway-enforced IFC for AI agents — 138 Kani proofs, adversarial evaluation
- Formal verification artifacts — TLA+ to Rust traceability
Apache-2.0 · Built in Rust