spendveto-sdk

Thin, dependency-free client for the SpendVeto spend-governance proxy. Agents never hold a wallet — they call these methods, and the proxy runs the full governance pipeline (freeze → policy → delegation caps → human approval) before signing anything.

npm install spendveto-sdk

import { SpendVeto, SpendVetoDenialError } from "spendveto-sdk";

const tg = new SpendVeto({ agentToken: "tg_..." }); // token from registerAgent()

// Pay for a governed tool — throws if any gate refuses.
try {
  const result = await tg.pay("review");
  console.log(result.settlement.receiptId);
} catch (e) {
  if (e instanceof SpendVetoDenialError) {
    console.log(e.code, "→", e.suggestion); // machine-readable, self-correctable
  }
}

// Preview a spend with zero side effects.
const preview = await tg.dryRun("summarize"); // { decision: "would_pause_for_approval", ... }

// Marketplace tools take per-call input via `query` — the one channel a paid GET-behind-x402
// call has for per-call data. Forwarded as query params onto the tool's own upstream request.
const matched = await tg.pay("basis-match", { query: { opinion: "rates fall this year" } });
console.log(matched.data.result.text);

// Governed LLM/API spend — estimated up front, fails closed if over budget.
const reply = await tg.chat("summarize this doc…");

Every refusal is a typed SpendVetoDenialError carrying a machine-readable code and a concrete suggestion, so an agent adjusts instead of crashing or retry-looping.

Requires a running SpendVeto proxy (npm run proxy) and server (npm run server) — see the main repo.

Apache-2.0.