System Architecture

ZeroTrust Agents acts as a secure, transparent reverse proxy sitting between your autonomous AI agents and language model API providers.

Request Flow

1. Request Interception: Your agent client app sends chat completion requests to the Gateway instead of calling OpenAI/Anthropic directly.
2. Policy Evaluation: The Gateway parses the requested tools and parameters, checking them against policies loaded in the local cache or Redis storage.
3. Semantic DLP Inspection: For sensitive operations, a DLP engine inspects parameters for credential leaks, toxic prompts, or SQL injections using local embeddings or models.
4. Upstream Routing & Streaming: If policies approve, the Gateway forwards the request to the upstream API and streams back the completions transparently to your agent.

Storage & Caching

The control plane utilizes Redis for cluster-wide distributed rate limiting and policy configurations, ensuring rapid authorization checks under high load.