PRIVATE INFERENCE NETWORKCA PENDING$INFERBASE READY
Agent calls should not leak by default
Private AI inference.
00Compute routes encrypted model calls to decentralized GPU nodes, returns verifiable execution receipts, and settles usage through INFER.
00router/live
requestagent.wallet / llama-4-private
policyprivate + receipt required
selected laneTEE-07 / 184ms
receipt0x9ab0...verified
Inference market
Agents buy model calls like network work.
The router treats every request as a priced job with privacy policy, model target, node selection, execution metadata, and settlement.
ModelLaneNodeLatencyPrice
llama-4-privateTEE laneA100184ms0.018 INFER
qwen-agent-fastencryptedL40S96ms0.006 INFER
mixtral-toolspublic4090212ms0.004 INFER
deepseek-coderencryptedH100141ms0.014 INFER
Encrypted sessions
Agent prompts enter the router as private payloads, not public logs.
Node market
Operators publish capacity, model support, latency, and live pricing.
TEE lanes
Sensitive calls can route through isolated execution lanes when required.
Receipts
Every completed call returns model, node, cost, latency, and proof state.
Proof surface
Every inference call can leave a receipt.
Receipts make the network legible: which model answered, which node served it, what privacy lane ran, how much it cost, and whether proof passed.
Prompt hash, not prompt leak Node and model metadata Settlement asset and cost
{
"request_id": "req_agentwallet_0001",
"model_id": "llama-4-private",
"node_id": "node_tee_lane_07",
"privacy_lane": "tee",
"latency_ms": 184,
"cost": { "amount": "0.018", "asset": "INFER" },
"proof": { "status": "verified" }
}Launch packet
Built for the private inference lane.
The brand stays narrow: not a chatbot, not generic GPU hosting, but private inference infrastructure for agent wallets and onchain apps.
Name00Compute
Handle@00computeAI
Ticker$INFER
CApending deploy
Website00compute.ai target