Cursor's agents now record videos. Here's how to add the same capability to your own AI agents

Three days ago, Cursor announced that their AI agents can now record their work — videos, logs, and screenshots of every action they take. CNBC picked it up. The message was clear: developers want agents that can show their work, not just describe it.

But here's the catch: Cursor built this capability for themselves. It's locked inside their product. If you're building your own AI agents with Claude, LangChain, CrewAI, or any other framework, you're out of luck.

Until now.

The problem Cursor solved (for themselves)

When an AI agent does work on the web, you have no proof it actually worked. Did it click the button? Did the button do what you expected? Is the agent lying or hallucinating?

Video is proof. A 30-second recording of the agent opening the login page, filling credentials, navigating to the dashboard, and capturing a screenshot is unambiguous. You can see exactly what happened.

Cursor shipped this because they realized their agents needed to prove their actions. It's the most important debugging feature an agent framework can have.

The solution: PageBolt's record_video endpoint

PageBolt's record_video endpoint does exactly what Cursor built, but as an open API. Call it from any agent, any framework, any runtime.

Here's how to add video recording to a LangChain agent in 40 lines of code:

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic();
const PAGEBOLT_API_KEY = process.env.PAGEBOLT_API_KEY;

async function recordAgentWorkflow() {
  // Define the agent's workflow as browser steps
  const steps = [
    { action: "navigate", url: "https://example.com/dashboard" },
    { action: "wait", ms: 2000 },
    { action: "click", selector: "button[data-action='export']" },
    { action: "wait", ms: 3000 },
    { action: "scroll", y: 300 }
  ];

  // Record the entire workflow as a video
  const response = await fetch("https://pagebolt.dev/api/v1/record_video", {
    method: "POST",
    headers: {
      "x-api-key": PAGEBOLT_API_KEY,
      "Content-Type": "application/json"
    },
    body: JSON.stringify({
      steps: steps,
      pace: "normal",
      format: "mp4",
      cursor: { visible: true, style: "highlight" },
      frame: { enabled: true, style: "macos" }
    })
  });

  const result = await response.json();
  console.log("Video recorded:", result.output_file);
  return result.output_file;
}

recordAgentWorkflow();

That's it. One API call. The agent's entire workflow is now a professional-quality video with cursor tracking, click effects, and browser chrome.

Why agents need video recording

Debugging: When an agent fails, you don't get a stack trace — you get a confused state. Video shows you exactly where it went wrong.

Proof of work: For financial transactions, contract signing, or high-stakes automation, video is compliance. "The agent clicked the confirm button" + video = proof.

Stakeholder demos: Instead of describing what your agent does, show them. 60 seconds of video beats 10 minutes of explanation.

Agent evaluation: Teams benchmarking agent frameworks (Claude vs GPT-4 vs local models) need reproducible evidence. Video is that evidence.

Real-world example: CrewAI agent with video output

import Anthropic from "@anthropic-ai/sdk";

// CrewAI agent that books a meeting
async function crewaiAgentWithVideo() {
  const agent = new Anthropic({
    apiKey: process.env.ANTHROPIC_API_KEY,
  });

  // Agent records itself booking a meeting
  const videoSteps = [
    { action: "navigate", url: "https://calendly.com/demo" },
    { action: "wait", ms: 1500 },
    { action: "click", selector: "input[type='email']" },
    { action: "fill", selector: "input[type='email']", value: "[e-mail]" },
    { action: "scroll", y: 400 }
  ];

  // Record the booking workflow
  const video = await fetch("https://pagebolt.dev/api/v1/record_video", {
    method: "POST",
    headers: {
      "x-api-key": process.env.PAGEBOLT_API_KEY,
      "Content-Type": "application/json"
    },
    body: JSON.stringify({
      steps: videoSteps,
      pace: "normal",
      format: "mp4"
    })
  }).then(r => r.json());

  return {
    success: true,
    video: video.output_file,
    message: "Meeting booked and recorded"
  };
}

Why Cursor's announcement matters to you

Cursor has 50K+ paid users and unlimited dev tooling budget. If they think agents need video recording, they're right. But you shouldn't have to rebuild it yourself.

PageBolt's record_video endpoint is:

• Fast: 2-5 seconds for a typical workflow
• Reliable: Handles dynamic content, authentication, JavaScript-heavy sites
• Reusable: Same endpoint works for any agent, any framework
• Documented: Full API docs at pagebolt.dev/docs

Next steps

1. Copy the code example above into your agent
2. Replace the workflow steps with your own
3. Call /record_video after your agent finishes
4. Share the MP4 with your team

You just shipped what Cursor spent months building.

Try it free: 100 requests/month, no credit card. Get started →

---

PageBolt is the open API for agent video recording. Use it to add the same capability to Claude, LangChain, CrewAI, or any agent framework.