TrustedRouter is a privacy-focused AI gateway designed to provide developers with a single, OpenAI-compatible API that bridges access to hundreds of open-weight and frontier AI models. Created to address the needs of teams requiring verifiable privacy, the platform offers a robust architecture where requests are routed through an attested gateway. This gateway is specifically engineered to ensure that no prompt or output content is logged, with architecture designed so that not even the company's own engineers can access user requests. By providing a unified interface, TrustedRouter allows users to switch models or providers without changing their code base, requiring only a simple base URL modification.

Functionality centers on acting as a secure intermediary between developers and various AI model providers. It implements intelligent routing policies, including provider failover, regional routing, and zero-retention paths. By maintaining an attested prompt path, TrustedRouter ensures that requests are handled securely, with transparency provided through public status pages and verifiable trust evidence that links the running gateway to open-source code commits.

Some of the key features are:

• Attested Gateway: The production prompt path runs in a secure environment with published, verifiable evidence that connects the gateway image to public source code.
• Zero-Retention Routing: Dedicated routes ensure requests are only sent to providers that contractually commit to zero data retention.
• Provider Failover: Automatic, intelligent routing to healthy providers prevents single-provider outages from impacting the overall product.
• Confidential Compute & E2EE: Support for routes that leverage provider-side confidential compute and end-to-end encryption for maximum data security.
• OpenAI-Compatible API: Maintaining compatibility with standard OpenAI SDKs for easy integration and rapid migration.
• Regional Routing: Ability to pin traffic to specific geographic regions, such as EU-focused endpoints, for compliance and latency management.
• BYOK Support: Capability to bring your own provider keys, allowing teams to utilize existing enterprise rate limits or commitments.

TrustedRouter is designed for ease of use, requiring only a simple change to the base URL in existing OpenAI-compatible calls. It offers specific routing aliases, such as trustedrouter/auto for general traffic, trustedrouter/zdr for sensitive legal or professional work, and trustedrouter/e2e for confidential workloads. Developers can also use the model advisor skill to automate model selection based on cost, latency, AI IQ, and privacy requirements.

Some common use cases include:

• Legal and Sensitive Data Processing: Using zero-retention routes to ensure that proprietary legal documents or sensitive data are never stored by AI providers.
• High-Reliability Production Systems: Leveraging automated provider failover to ensure consistent AI model availability for user-facing applications.
• European Compliance: Routing European user data through specific EU-based infrastructure to meet regional data residency requirements.
• Cost-Optimized Development: Combining frontier and open-source models through a single API to achieve the best performance-to-cost ratio for different tasks.