Building the trust layer for autonomous AI
ProvnAI builds security infrastructure for autonomous AI agents. The goal is simple: powerful agent systems should be secure, reviewable, and accountable where they touch the real world.
Why this exists
AI agents are gaining access to tools, filesystems, APIs, and operational systems faster than the surrounding control layers are maturing.
In too many deployments, the actual security model is still a request embedded in a prompt. That is not enough for systems that can act on real systems.
ProvnAI focuses on the layer between model reasoning and privileged execution: the place where policy should be enforced, evidence should be generated, and trust should become inspectable.
McpVanguard
Security proxy for MCP and agent tooling, with deterministic inspection and enforcement before privileged tool actions execute.
VEX Protocol
A governance model for wrapping actions that matter in evidence that can be reviewed, verified, and enforced at the execution boundary outside the model.
Evidence Workflows
Practical review surfaces and audit patterns that make governed execution usable for security, platform, and compliance teams.
Team
ProvnAI is building infrastructure for autonomous AI security and governance.
History
The project has moved from protocol framing into a more disciplined product and architecture surface over the last two release cycles.
Initial protocol work begins around execution-boundary governance for autonomous AI.
McpVanguard reaches first public release on PyPI as a security proxy for MCP.
Early protocol work around Evidence Capsules and governed execution matures into a clearer architecture internally.
Verification and audit workflow prototypes are formalized for browser-facing review flows.
The public website and documentation are consolidated around a clearer product and architecture story.
Open where it matters
McpVanguard is open source and MIT licensed. Governance and protocol work is developed with design partners and released where it is ready to be inspectable and useful to the broader ecosystem.