Prediction Guard

Prediction Guard is a self hosted control plane for AI agents, deployed inside the customer's own cloud account or security boundary. It sits between applications and the model, tool and compute layers, and is built on zero trust principles to identify, authorize, enforce, meter, record and contain the actions an agent takes. It is aimed at platform engineers, AI engineers and software vendors who need to run agents in regulated or compliance constrained environments. The platform includes an agent supply chain registry for models, tools and MCP servers, an OpenAI and Anthropic compatible API gateway, and runtime controls scoped per API key. Input and output safeguards cover PII masking, prompt injection and toxicity checks. An immutable audit log and usage telemetry support investigation, and a no code builder called Agent Forge, along with an API and SDK, lets teams build and deploy agents. Its policies are aligned with NIST and OWASP guidance, and the vendor publishes an Agent Control Kit on Docker Hub as a verified publisher. There are two ways to buy. The regular tier is billed through the AWS or Azure marketplace, with current rates shown on each marketplace listing, and includes the self hosted control plane, Agent Forge, documentation and email support. The Enterprise tier is custom priced through annual agreements and adds single tenant or air gapped deployment, SSO and SAML, role based access control and priority support with a contracted SLA. Public sector buyers can purchase through Carahsoft.