Skip to main content

Vercel AI SDK Integration

Use @sequenceware/vercel-ai to enforce Sequenceware policies on every Vercel AI SDK tool execution.

1. Install packages​

npm install @sequenceware/sdk @sequenceware/vercel-ai ai

2. Create an Sequenceware client and run​

import { SequencewareClient } from '@sequenceware/sdk';

const client = new SequencewareClient({
  baseUrl: 'http://localhost:3000',
  apiKey: process.env.SEQUENCEWARE_API_KEY!,
  agentId: 'vercel-ai-agent',
});

const runId = await client.startRun(undefined, {
  source: 'vercel-ai',
  task: 'governed tool execution',
});

3. Wrap tools with sequencewareMiddleware​

import { generateText, tool } from 'ai';
import { openai } from '@ai-sdk/openai';
import { sequencewareMiddleware, isSequencewareBlockingError } from '@sequenceware/vercel-ai';
import { z } from 'zod';

try {
  const result = await generateText({
    model: openai('gpt-4.1'),
    prompt: 'Create a deployment plan and store it in infra/plan.md',
    ...sequencewareMiddleware({
      client,
      runId,
      tools: {
        write_file: tool({
          inputSchema: z.object({
            path: z.string(),
            content: z.string(),
          }),
          execute: async ({ path, content }) => {
            // perform the side effect
            return { ok: true, path, bytes: content.length };
          },
        }),
      },
    }),
  });

  console.log(result.text);
} catch (error) {
  if (isSequencewareBlockingError(error)) {
    console.error('Blocked by Sequenceware:', error.message);
  } else {
    throw error;
  }
} finally {
  await client.completeRun(runId);
}

Behavior​

  • allowed / warn: tool executes and result is tracked.
  • blocked: throws SequencewareToolBlockedError.
  • awaiting_approval: throws SequencewareApprovalRequiredError.

Notes​

  • The integration expects an explicit runId; it does not auto-manage runs.
  • By default it fails open if Sequenceware is unreachable (onCheckError: 'allow').
  • Set onCheckError: 'block' to fail closed.