@sip402/splitter

· repo · SPEC

A reference seller built on sip402: resell Venice AI inference behind an OpenAI-compatible gateway, billed per token as a paid stream — draws settle on-chain in batches while the response streams. Shows the full seller loop end to end.

npm i @sip402/splitter

What's inside

| Export | Purpose | |---|---| | veniceUpstream(privateKey) / localUpstream() | The AI upstream. veniceUpstream is real Venice via x402 (Base mainnet only); localUpstream is a deterministic generator for testnet/dev. | | tokenCostAtoms(), USDC_PER_1K_TOKENS, ATOMS_PER_TOKEN | Per-token USDC pricing (integer atoms, ceil-rounded). | | StreamingDrawer | Accrue token costs and fire an on-chain draw each time the batch threshold is crossed; DryTabError when the cap is hit. | | makeGateway() | A Hono app that wraps an upstream with sip402 metering + a settlement bus + the seller address. |

Stream + draw

import { StreamingDrawer, veniceUpstream, localUpstream, tokenCostAtoms } from "@sip402/splitter";

const drawer = new StreamingDrawer({ sellerPrivateKey, commitment, minBatchAtoms, onEvent });
const up = IS_MAINNET ? veniceUpstream(PRIVATE_KEY) : localUpstream();

for await (const chunk of up.chatStream({ model: "llama-3.3-70b", messages })) {
  process.stdout.write(chunk.text);
  await drawer.record(tokenCostAtoms(chunk.tokens)); // draws on-chain at each threshold
}
await drawer.finalize(); // flush the remainder; throws DryTabError if the cap is dry

OpenAI-compatible gateway

import { serve } from "@hono/node-server";
import { makeGateway, veniceUpstream } from "@sip402/splitter";

const { app, bus, sellerAddress } = makeGateway({
  sellerPrivateKey: SELLER_KEY,
  upstream: veniceUpstream(PRIVATE_KEY),
  minBatchAtoms: toUsdcAtoms("0.25"),
});
serve(app); // POST /v1/chat/completions — pay-per-token over sip402

Related

• @sip402/server — the verify/accumulate/batch-redeem engine it builds on.
• @sip402/core · @sip402/client

License

MIT