Comparing private OpenAI-compatible LLM APIs: zero retention, TEEs and verifiable enclaves
A new comparison reviews OpenAI-compatible LLM APIs that promise privacy for hosted model requests. It distinguishes three approaches: zero data retention, which relies on a provider's contractual promise, trusted execution environments that isolate models in hardware, and end-to-end encryption to a key held only by the enclave. The author argues the key question is whether encryption ends at a gateway or at the enclave running the model, and points to attestation as a way to verify claims against published code. The comparison covers providers including Morse, OpenRouter, TrustedRouter, Phala, NEAR AI Cloud, Tinfoil, Prem and Privatemode. The author notes that "not described" entries mean only that a feature could not be found on a company's own pages, not that it is absent. The data was checked against provider documentation on November 20, 2026.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in