Spend controls for Vercel AI SDK tools

The Vercel AI SDK runs tools through generateText / streamText with optional toolApproval. Paybond supplies that approval hook and wraps execute on side-effecting tools — spend verify before side effects, auto-evidence after success. Keep your provider client for inference; Paybond guards paid tool calls only.

TypeScript only. paybond-kit Python rejects framework: "vercel-ai" (native adapter is @paybond/kit/vercel-ai). Use agent-agnostic spend controls for Python hosts.

Adapter reference: /docs/kit/vercel-ai.

• toolApproval with wrapped execute

  Harbor authorize runs before side-effecting tools; generateText/streamText stay on your provider.

• Auto-evidence

  Wrapped execute finalizes spend and submits signed evidence after tool success.

• Policy budgets

  Versioned YAML caps per-call and intent spend — default-deny for paid tools.

• TypeScript only

  Native @paybond/kit/vercel-ai adapter; Python hosts use agent-agnostic or another dual-language adapter.

Why Paybond (not just AI SDK guardrails)?

Vercel AI SDK tool approvals do not enforce a spend limit, a per-operation permission check (capability token), or a signed completion receipt tied to a spend agreement (intent).

How it works

3-minute quickstart

Smoke the vercel-ai adapter sandbox contract — no Vercel credentials required for this check:

paybond login
paybond agent demo vercel-ai smoke \
  --operation paid-tool \
  --requested-spend-cents 100 \
  --evidence-preset cost_and_completion \
  --format table

paybond login
paybond agent demo vercel-ai smoke \
  --operation paid-tool \
  --requested-spend-cents 100 \
  --evidence-preset cost_and_completion \
  --format table

When the smoke succeeds you should see:

• ✓ Spend approved
• ✓ Tool completed
• ✓ Evidence verified (cost_and_completion)

What success looks like

Scaffold

paybond init agent-middleware --framework vercel-ai --out paybond-vercel-ai.ts
paybond policy init --preset travel --out paybond.policy.yaml

paybond init agent-middleware --framework vercel-ai --out paybond-vercel-ai.ts
paybond policy init --preset travel --out paybond.policy.yaml

Validate before deploy:

paybond policy validate-tools --file paybond.policy.yaml --local-only

paybond policy validate-tools --file paybond.policy.yaml --local-only

Wire middleware

import { generateText, tool } from "ai";
import { openai } from "@ai-sdk/openai";
import { z } from "zod";
import { Paybond } from "@paybond/kit";

const paybond = await Paybond.open({ apiKey: process.env.PAYBOND_API_KEY! });

const { agentTools: tools, toolApproval } = await paybond.agent({
  policy: "travel",
  framework: "vercel-ai",
  tools: {
    bookHotel: tool({
      description: "Book a hotel room",
      inputSchema: z.object({
        city: z.string(),
        estimatedPriceCents: z.number().int().nonnegative(),
      }),
      execute: async (args) => bookHotel(args),
    }),
    searchWeb: tool({
      description: "Search the web",
      inputSchema: z.object({ query: z.string() }),
      execute: async (args) => searchWeb(args),
    }),
  },
});

const result = await generateText({
  model: openai("gpt-4.1"),
  tools,
  toolApproval,
  prompt: "Find hotels in Lisbon and book one under budget.",
});

Already bound a run? paybond.wrapTools(run, tools, { framework: "vercel-ai" }) returns { tools, toolApproval } without reloading policy.

For a full App Router example with request-scoped bind, see Next.js agent checkout.

Production: omit sandbox defaults, use deferred bind, then instrumented.bind({ intentId, capabilityToken }) per session.

Production checklist

Works with

Ready to test?

Related guides

• Next.js agent checkout — App Router and Vercel AI
• Agent middleware — lazy context and tenant isolation
• Let agents buy groceries — shopping preset example

Developer reference: /docs/kit/vercel-ai.