TrustedRouter

TrustedRouter is an end-to-end encrypted AI gateway designed for routing large language model (LLM) traffic across multiple providers, aimed at teams building production AI applications where privacy, reliability, and verifiability are critical. It offers a single OpenAI-compatible API that enables users to send requests to a wide range of models, including open-weight options such as Qwen, GLM, DeepSeek, Gemma, Kimi, and MiniMax, as well as other frontier models. The platform is built to address challenges such as provider downtime, model quality changes, cost fluctuations, and the need for privacy controls and verifiable trust paths.

A key feature of TrustedRouter is its attested gateway architecture, which ensures that prompt and output data are not logged or stored in the control plane. Only metadata necessary for billing and operations is retained, with prompt and output bodies excluded from logs. The gateway publishes attestation evidence, source commit, and image digest, allowing technical customers to verify the exact code and environment running the router, rather than relying on opaque promises. This transparency is supported by a public trust page and open-source backend and configuration.

The service supports provider failover, EU regional routing, and a preference for zero-retention providers, allowing teams to specify data collection policies per provider. TrustedRouter also offers BYOK (bring your own key) and prepaid credits for usage, and is compatible with existing OpenAI SDKs, making migration straightforward by changing the base URL. The platform is available across multiple regions, including Iowa, Netherlands, N. Virginia, Tokyo, Hong Kong, Singapore, and São Paulo, with three live regions and edge capacity warming to support reliable global routing. 99% uptime.

TrustedRouter is positioned as an open-source alternative to closed AI routers, particularly for teams in sectors like legal, finance, and production AI who require strict privacy, resilience against provider failures, and verifiable control over inference paths. The platform does not log prompts or outputs, supports public attestation, and enables more resilient and trustworthy AI product deployments.