About

OffRouter is a private AI inference service for software agents and developers. It provides an OpenAI-compatible API, so an existing application can switch to it by changing the base URL. The models it serves, 21 open models including DeepSeek, Qwen, Kimi, GLM, Gemma and Llama, run inside confidential GPU enclaves (Intel TDX with NVIDIA confidential computing), and prompts and answers are not stored or logged. Each response includes a receipt signed inside the enclave that can be checked against the enclave's hardware attestation. OffRouter does not require an account.