General

Which models does Panda run?

Panda run DeepSeek R1 0528 & Llama 4 scout. Both are full & quantized models.

Where is Panda’s server located?

It’s located in Norway.

How does Panda protect users’ privacy?

Panda runs every conversation inside an Intel TDX & NVIDIA Confidential Computing enclave, keeps the TLS key inside that enclave, and never stores your prompts. So nobody—not even Panda admins—can read your data before, during, or after inference.

• Deep-dive

Inference

How does inference work at a high level?

How does Panda’s end-to-end-encrypted inference work?

A fresh TLS key is born inside the enclave, baked into an Intel-signed attestation, and proven to your client on every response—so encryption really runs from your browser straight to the model process and nowhere else.

• Deep-dive

How can I verify that my conversation was end-to-end encrypted?

Your Panda client automatically does this for you, but you can also do this manually. Grab the two Panda-* headers from any reply, fetch the matching attestation token, and verify three signatures—Intel’s JWT, the quote’s pub-key binding, and the response signature.

• Deep-dive

How does Intel TDX + NVIDIA Confidential Computing guarantee the inference is private?

TDX keeps every CPU byte inside an encrypted Confidential VM, NVIDIA’s Hopper/Blackwell GPUs keep every GPU byte inside an encrypted Compute-Protected Region (CPR), and Panda binds the two attestation proofs together so the client can verify cryptographically that both were enabled for every single reply.