Inference
For private model execution
Attest
/01 PROTOCOL LAYER- ✓ Composite CPU + GPU quotes
- ✓ Signed serving images
- ✓ Encrypted prompt delivery
- ✓ Per-response receipts
Confidential AI for the institutional chain.
Enclave runs AI inside attested hardware. Keep prompts private, verify every receipt, and settle agent inference in USDC.
/
PRIVATE IN. PROVABLE OUT.
Keep your intelligence
inside the enclave
/
THE PROTOCOL
Confidential by design
From an encrypted prompt to a verifiable receipt, Enclave brings private compute and agent commerce into one flow.
Sealed agents
Keep model, memory and policy inside the enclave, with a wallet and an enforced spending mandate.
Private USDC payments
Settle per inference through x402. Confidential transfers shield amounts while addresses remain visible.
Selective disclosure
Give auditors scoped view keys and receipt exports without opening private prompts or model outputs.
/
TECHNOLOGY
Private computation. Proven.
/
THE INFERENCE LIFECYCLE
/PROTOCOL ASSURANCES
No secret is released until the hardware attestation verifies. Trust is checked before the first token.
Attestation first
THE ENCLAVE SECURITY MODEL/
THE FOUNDATION
Built on trusted hardware
/01
GPU
NVIDIA CC
/02
TEE
CPU TEE
/03
ARC
Arc
/04
USDC
USDC
/05
x402
x402
/THE ROADMAP
Inside the
Enclave roadmap
A staged path from private inference to sealed agent commerce and frontier-scale compute.
Explore the architecture↗Built in stages.
Milestones describe the product roadmap. Availability is announced as each phase is ready.
E1Private inference
+
Encrypted requests, composite attestation, signed inference receipts, Arc anchoring and USDC settlement.
Explore this stage ↗E2Sealed agents
+
An enclave runtime for model, memory and policy, with wallets, on-chain mandates and private payments.
Explore this stage ↗E3Model marketplace
+
Attested model listings, provider stakes, fee accounting, receipt exports and scoped compliance tools.
Explore this stage ↗E4Frontier compute
+
Multi-GPU Blackwell clusters and fine-tuning inside the enclave extend the confidential execution boundary.
Explore this stage ↗/PROTOCOL LAYERS
One private stack.
COMPUTE → PROOF → PAYMENTCommerce
For agents that transact
Settle
/02 PROTOCOL LAYER- ✓ Per-call USDC accounting
- ✓ x402 / Nanopayments
- ✓ Enforced spending mandates
- ✓ Shielded payment amounts
Compliance
For verifiable oversight
Verify
/03 PROTOCOL LAYER- ✓ Model and code provenance
- ✓ Versioned measurement policy
- ✓ Scoped auditor view keys
- ✓ Receipt exports
/FAQ
Frequently
Asked
Questions
What does Enclave keep private?+
The intended execution path keeps prompts, model memory and outputs inside an attested CPU and GPU environment. Requests are encrypted, TLS terminates in the confidential virtual machine, and operational records contain hashes rather than plaintext inputs or outputs.
What does an inference receipt prove?+
A receipt records the model hash, code hash, input and output hashes, attestation reference, timestamp and signature. It provides provenance for an approved execution environment. It does not prove that the model’s answer is correct.
How are inference calls paid for?+
The protocol is designed for per-call USDC settlement through x402 or Nanopayments. Arc confidential transfers shield amounts; addresses remain visible. Agent mandates set spending limits.
What are sealed agents?+
Sealed agents keep their model, memory and policy inside the enclave. The planned runtime connects these agents to a wallet, spending mandates and signed receipts for their actions.
Is this dashboard connected to live infrastructure?+
This is an interactive product preview. It uses sample models and simulated usage, receipts, payments and agents. No GPU inference is performed, no funds move, and no receipt is anchored on a blockchain.
/INSIGHTS
Inside private AI
More insights↗
Attestation before inference: where privacy begins
Why a CPU and GPU quote must verify before keys, prompts or model weights are released.
A receipt for every answer. No prompt disclosure.
The anatomy of a signed inference receipt and the limits of what it proves.
Private AI meets programmable USDC payments.
How per-call settlement and mandates fit into the sealed agent lifecycle.