PRIVATE INFERENCE NETWORKCA PENDING$INFERBASE READY
Infrastructure for machine-to-machine inference
Private AI
inference.
00Compute is a permissionless routing layer that connects autonomous agents to encrypted GPU inference. No prompt logging. No data exhaust. Just signed calls, verifiable receipts, and settlement in INFER.
00router/live
requestagent.wallet / llama-4-private
policyprivate + receipt required
selected laneTEE-07 / 184ms
receipt0x9ab0...verified
Network model
Four layers. One call path.
Every inference request moves through four deterministic layers. No routing magic — just policy, matching, execution, and proof.
01
Agent walletOriginates signed inference request with privacy policy attached
02
RouterMatches request to available nodes by lane, latency, model support, and price
03
GPU nodeExecutes inference inside the selected privacy envelope; cannot read prompt plaintext in TEE mode
04
SettlementReceipt is written, INFER settlement is triggered, proof hash is returned to caller
Inference market
Agents buy model calls like network work.
The router treats every request as a priced job: model target, privacy policy, node selection, execution metadata, and settlement. Agents shop by policy, not by vendor.
ModelLaneNodeLatencyPrice
llama-4-privateTEE laneA100184ms0.018 INFER
qwen-agent-fastencryptedL40S96ms0.006 INFER
mixtral-toolspublic4090212ms0.004 INFER
deepseek-coderencryptedH100141ms0.014 INFER
Encrypted sessions
Agent prompts enter the router as private payloads. No prompt logging, no data exhaust, no inference surveillance.
Node market
GPU operators publish capacity, model support, latency profiles, and live pricing. Agents shop by policy, not by brand.
TEE execution lanes
Sensitive inference runs inside hardware-isolated enclaves. The node cannot inspect the payload even while serving.
Verifiable receipts
Every call returns model identity, node metadata, privacy lane, latency, cost in INFER, and proof status.
Why this layer exists
Autonomous agents are not chat users.
They need infrastructure that treats inference as network work: priced, private, verifiable, and settlement-ready. 00Compute is that substrate.
Agents are API consumers
They don't need chat UX. They need signed inference calls with policy metadata, priced lanes, and verifiable receipts.
Model calls are network work
Every inference is a job: a model target, a privacy lane, a node, a cost, and a proof. The router treats them as such.
Privacy is not optional
Autonomous agents carry wallet state, strategy context, and private keys. Their inference traffic must be as opaque as their transactions.
Proof surface
Every inference call leaves a receipt.
Receipts make the network legible: which model answered, which node served, what privacy lane ran, how much it cost in INFER, and whether proof passed. No black boxes. No untraceable calls.
Prompt hash published, not prompt plaintext Node identity, model version, and lane metadata Settlement amount in INFER with proof status
{
"request_id": "req_agentwallet_0001",
"model_id": "llama-4-private",
"node_id": "node_tee_lane_07",
"privacy_lane": "tee",
"latency_ms": 184,
"cost": { "amount": "0.018", "asset": "INFER" },
"proof": { "status": "verified" }
}Execution flow
From request to settlement.
Four deterministic stages. Every inference call follows the same path.
Launch packet
Private inference is not a feature. It is the substrate.
00Compute stays narrow: not a chatbot, not generic GPU hosting, not a model marketplace. Just encrypted inference routing for autonomous agents that need privacy, proof, and settlement by default.
Name00Compute
Handle@00computeAI
Ticker$INFER
CApending deploy
Website00compute.ai target