Kickstart live

Backers open

Inference checkout

Stream the selected model route.

Pay the selected CPX/request ask, then receive an OpenAI-compatible stream.

Quote0.015 CPX/request
Latency14ms
DeliverySSE stream
Router decision: ranked by price, latency, reliability, and remaining capacity. Selected Meta Llama-3-70B-Instruct at 0.015 CPX/request with 14ms latency. Tokens Streamed: 0 | Target Seller Route: Meta Llama-3-70B-Instruct | Settlement Status: 0.015 CPX required.

Active surplus capacity registry

Selectable route book

3 routes / static-seed

Meta Llama-3-70B-Instruct
Stream outputMeta Llama-3-70B-Instruct / 14ms
Awaiting route-specific CPX settlement and SSE stream...
Tokens Streamed: 0 | Target Seller Route: Meta Llama-3-70B-Instruct | Settlement Status: 0.015 CPX required.
Developer details

OpenAI-compatible proxy

import OpenAI from 'openai';

const capixClient = new OpenAI({
  apiKey: "cpx_live_market_token_session_key",
  baseURL: "https://capix.network/api/v1"
});

await capixClient.chat.completions.create({
  model: "meta-llama-3-70b-instruct",
  messages: [{ role: "user", content: "Route this through CapIX." }],
  stream: true
}, {
  headers: {
    "x-cpx-route-id": "meta-llama-3-70b-instruct",
    "x-cpx-wallet-address": "<buyer-wallet>",
    "x-cpx-settlement-signature": "<confirmed-cpx-tx>"
  }
});

Inference market primitives

SupplySurplus inference capacityModel endpoints, GPU clusters, and reserved provider windows can be listed as sellable routes.
DemandApps, agents, evals, batch jobsBuyers route requests through the CPX gateway when they want cheaper or overflow inference.
PriceDiscounted tokens / request creditsRoutes can clear by per-request CPX settlement, per-token credit, or reserved batch window.
RouterSettle -> validate -> dispatchThe API validates the settlement signature, then dispatches to local Ollama, a standby external route, or the next seller.
Seller flow, buyer flow, and hackathon audit

Seller endpoint flow

  1. 01Register an OpenAI-compatible endpoint, local Ollama node, GPU cluster, or reserved provider capacity window that matches open demand.
  2. 02Advertise price, latency target, reliability, context length, available request budget, and payout wallet.
  3. 03Run a health check so the router can rank your route by price, latency, reliability, and capacity.
  4. 04When buyer requests route through your endpoint, CapIX records the CPX settlement lane against your listing.
Base URLModel IDBearer tokenContext lengthMax requests/minPayout wallet

Buyer route flow

  1. 01Select a model route by price, latency, reliability, and endpoint type.
  2. 02Spend CPX per request or use credits against a reserved batch window.
  3. 03CapIX validates the settlement on-chain before forwarding the request to the selected seller route.
  4. 04The response streams back through the same OpenAI-compatible gateway with settlement proof attached.
Seller endpointCluster #7041
Buyer route0.015 CPX/request
Market spreadDiscounted below incumbent APIs

Track 02 audit

RequirementProofSurface
Marketplace, router, exchange, or tokenCPX-settled inference router/sell?track=inference
Seller listing flowsEndpoint and cluster manifest/inference
Buyer route selectionSelectable model route book/inference
SupplySurplus inference capacity/inference
DemandApps, agents, evals, batch jobs/sell
PriceDiscounted CPX request market/inference