Terms of Service & Privacy Policy

01 · Scope

Who these terms cover

These terms govern access to and use of the Ghost Compute substrate, the Ghost Worker desktop client, the Inference Console, the Confidential Dark Pool, the Attestation Explorer, the Solana Anchor programs, and the GHST token. By running a Ghost, submitting a compute job, or holding GHST, you accept these terms.

Ghost Compute fuses two missing pieces: compute networks lack privacy; privacy protocols lack compute. Everything below follows from that thesis.

02 · The Substrate

Confidential GPU network

The substrate is a permissionless network of confidential GPUs — NVIDIA H100 / H200 Confidential Compute and AMD SEV-SNP enclaves — coordinated by six Solana Anchor programs. Every job is bound to a hardware attestation. A broken attestation halts the job rather than leaking plaintext: the substrate fails closed by design.

Five subsystems

• Confidential Worker runtime (Tauri-based Ghost Worker client).
• Layered Privacy Envelope: TEE · ZK · MPC · FHE.
• Private Inference via x402 pay-per-call.
• Dark-Pool Engine with TEE/FHE matching and Jito-bundle settlement.
• Six Solana Anchor programs forming the onchain control plane.

03 · Operators (Ghosts)

Running a Ghost

Operators contribute confidential GPU capacity by running Ghost Worker. The client detects supported TEEs, registers the node, signs hardware attestations, executes sealed jobs, and receives payouts through Confidential Balances.

• Operators must run unmodified, audited builds of Ghost Worker.
• Tampering with attestation, the envelope, or the runtime forfeits stake and rewards.
• Operators do not observe job plaintext, prompts, weights, or order flow at any time.
• Earnings are paid in GHST via Confidential Balances and may be subject to staking multipliers.

04 · Compute Consumers

Submitting confidential jobs

Consumers — hedge funds, AI agents, enterprises, regulated industries, and privacy protocols — submit encrypted workloads to the substrate. Two primary surfaces are exposed:

• Private Inference — vLLM-style API with x402 pay-per-call. Prompt, system message, memory, and tool calls remain sealed inside the enclave.
• Confidential Dark Pool — encrypted order submission, TEE/FHE matching, MEV-proof execution, confidential settlement via Jito bundles.

Consumers are responsible for the lawfulness of the workloads they submit. Use of the substrate to violate sanctions, target critical infrastructure, or process material the consumer has no right to process is prohibited.

05 · GHST Token

Token economy

GHST is an SPL Token-2022 asset with a TransferHook fee collector. Confidential Balances shield operator payouts. Fee routing distributes value as follows:

• 60% to stakers
• 20% to workers
• 10% burned
• 10% to treasury

GHST is a utility token used to pay for confidential compute, dark-pool execution, and governance. It is not a security, deposit, or claim on Ghost Compute revenue. Markets, prices, and liquidity are provided by third parties and are not warranted by Ghost Compute.

06 · Privacy Envelope

What stays sealed

The Privacy Envelope layers four complementary primitives — TEE for hardware isolation, ZK for verifiable computation without disclosure, MPC for distributed key management, and FHE for matching over encrypted state. The result: plaintext is unreachable to operators, validators, Ghost Compute, and any third party.

• We do not observe prompts, weights, order books, or settlement deltas.
• We cannot decrypt data sealed to a hardware attestation we did not issue.
• We publish attestation digests, audit reports, and the source of every audited release.

07 · Attestation & Audit

Multi-proof verification

Every confidential job is bound to a hardware attestation verified onchain. Multi-proof verification (TEE quote + ZK proof + onchain anchor) is published to the Attestation Explorer. Audits are performed by independent firms:

• Substrate & Private Inference — OtterSec
• Confidential Dark Pool — Trail of Bits and Zellic
• GHST · Token-2022 & settlement — Halborn

08 · Liability & Fail-Closed

Fails closed, not open

The substrate is designed to fail closed: a broken attestation halts the job rather than leaking plaintext. Ghost Compute does not warrant uptime, throughput, market outcomes, or token price, and is not liable for indirect, incidental, or consequential damages arising from use of the substrate. The substrate is provided “as-is” to the extent permitted by applicable law.

09 · Data Retention

What we keep, what we cannot keep

Because plaintext never leaves the enclave, Ghost Compute has nothing to retain about your workloads. We retain only the public artifacts required to operate a verifiable network:

• Attestation digests, public proofs, and onchain anchors.
• Aggregate, non-identifying telemetry from Ghost Worker (version, TEE class, uptime).
• GHST transfer history visible on Solana — shielded amounts via Confidential Balances.

We do not retain prompts, weights, order books, settlement deltas, or any plaintext state from confidential jobs. We cannot produce data we never possessed.

10 · Governance

Onchain policy

Privacy parameters, supported enclaves, verification policy, and treasury allocation are governed onchain. Governance proposals are executed without revealing what was computed. Material changes to these terms are announced in the Attestation Explorer and through official channels at least 14 days before they take effect, except where shorter notice is required by law or to address a security incident.

11 · Updates & Contact

Staying current

These terms may be updated as the substrate evolves through the four-phase rollout (Substrate + Private Inference → Confidential Dark Pool → Token + Settlement → Explorer + Governance). The effective date at the top of this document reflects the current version. Continued use of the substrate after an update constitutes acceptance.

For governance, audit, or integration inquiries, follow Ghost Compute on X.