Inject Xpense into an Agent

Goal: give an agent the ability to spend — either by exposing xpense's tool registry, or by gating the agent's own tool calls through a payment resolver. xpense emits intents; it does not own the agent loop.

Option A — expose the capability registry

1. xpense.listCapabilities() → array of ToolDefinition (name, description, JSON inputSchema).
2. Register them with your framework (each has a JSON Schema for arg validation).
3. Route calls through xpense.invoke(name, input, origin).

const tools = xpense.listCapabilities(); // x402.pay, payment.set_default, ...
const result = await xpense.invoke("x402.pay", {
  accepts: { amount: "0.25", currency: "USDC", resource: "/api/inference" }
});

Option B — gate pending tool calls

1. buildInjection({ resolvePayment }) — your resolver decides per call.
2. The agent calls injection.onPendingTool(call) before executing; gets "allow" | "deny".

import { buildInjection } from "@xagent/xpense";

const injection = buildInjection({
  resolvePayment: async (call) => {
    const { intent } = await xpense.emit(/* derive from call.args */);
    return { intent, decision: "allow" };
  }
});

const decision = await injection.onPendingTool({ tool: "search", args: {} });

Reference: capabilities · buildInjection · tooling types.