I’m exploring a software layer—analogous to public/private-key crypto—so a user can converse with an LLM where prompts and responses remain unreadable to all intermediaries, including the model host. (I mean “cipher” in the cryptographic sense.)

Two cases: Open-weights model: ensure the operator still can’t read prompts/responses. Closed, hosted model: true E2EE so even the provider can’t inspect content.

Topics we can discuss: Best near-term path: TEEs with attestation, FHE/HE, MPC/split inference, PIR for retrieval, differential privacy, or hybrids? How to handle key exchange/rotation for forward secrecy? Practical performance/accuracy limits (e.g., non-linearities, KV-cache, streaming)? Minimal viable architecture and realistic threat model? Any prior art or teams you’d point me to?

Please DM if you are interested in working with me.