FIRST RUN
Free
First LoRA fine-tune, small dataset (≤ 100k examples).
PRIVATE ALPHA · CONFIDENTIAL FINE-TUNING · TARGON CONFIDENTIAL GPUs
Doctus fine-tunes inside a sealed hardware enclave on Intel TDX + NVIDIA H200 confidential compute. Your training data is decrypted only inside the room. The trained adapter is encrypted to a key only you hold. We hold ciphertext and nothing else.
The privacy is not a policy. It is a property of the hardware: Intel TDX seals memory and CPU state; NVIDIA Protected PCIe encrypts the bus. Inside the room, the model trains on plaintext. Outside the room, there is only ciphertext.
We sit outside the room. We do not hold the key. We have nothing to hand over because nothing was ever ours to hold. The math, not the marketing, makes this true.
THE TRAINING PIPELINE
Your data exists only as ciphertext on Doctus-operated systems. The plaintext is defined only inside the attested enclave.
The trained weights are encrypted to your wallet's public key inside the enclave, before egress. We never hold your private key.
Inference runs in a second confidential enclave with no plaintext logging. End-user prompts and completions are sealed at rest and in motion.
No subscriptions, no seat fees, no idle GPU charges. Settle in TAO at oracle rate.