.svg?dpl=dpl_2DvYkfFFPPzLGnrMLC9FaBYaog7H){kind=small}
.svg?dpl=dpl_2DvYkfFFPPzLGnrMLC9FaBYaog7H){kind=small}
Research across sites
Ask an agent to compare product details or collect information from several pages. Include starting URLs and the fields to return in your task. Use the step history to review the pages it visited.
Run a HyperAgent taskAI Agents
Run browser tasks from a prompt
Run HyperAgent, Browser Use, or a computer-use model in a managed browser. Specify the task and step limit, then read the result and step history. Reuse an existing session when the task needs its login or open tabs.
Research, form entry, and data collection
Use signed-in applications
Pass the ID of a session that is already logged in. The agent can open records, navigate settings, or fill forms. Keep the browser open if another task will follow.
Reuse a sessionReturn structured data
Browser Use accepts an output schema for fields such as names, prices, and URLs. Enable output validation to check the response against that schema before your application uses the data.
Define the outputReview an address change before submitting it
First, ask the agent to open the customer's account and read the current shipping address. Keep the browser open while your application shows the current and proposed addresses for review. The reviewer can also inspect the account page through Live View.
Your application handles approval. Once the reviewer approves, start a second task in the same session to enter and save the new address. Check the task's status and result before showing a success message, then close the browser.
Task limits, results, and session settings
All agent APIs support starting tasks, checking status, and retrieving results. Additional options vary by agent.
Task limits
Set the maximum number of steps a task may take. Some agents also support a limit on repeated failures.
Background jobs
Start a task, save its job ID, and poll for completion. Use startAndWait in the SDK to handle polling in one call.
Live View
Watch the agent's browser session. Enable user control when someone needs to enter information or interact with the page.
Structured results
Define an output schema for Browser Use and enable response validation against it.
Sensitive data placeholders
Pass sensitive values separately to Browser Use. The model sees placeholders, which are replaced with the actual values during browser actions.
Browser configuration
Configure stealth, proxies, and profiles for a new browser session. Pass a session ID to use a browser you already configured.
Session reuse
Set keepBrowserOpen to retain the browser after a task. Pass its session ID to the next task, and stop the session when finished.
Bring your own keys
Use your own credentials for supported model providers. The provider bills model usage to your account; Hyperbrowser bills browser time.
Run a browser task
Start a task
Specify the starting URL, actions to take, and information to return. Choose an agent and set maxSteps. Pass a session ID if the task needs an existing browser.
Check progress
Poll the job's status or wait with the SDK. Open Live View to inspect the current page. The agent may reach its step limit before completing the task.
Read the result
Check the final status and read the returned data. Use the step history to investigate failures. Stop any session you kept open once all tasks are finished.
Set HYPERBROWSER_API_KEY in your environment. These examples ask HyperAgent for the title and URL of the top Hacker News story, with a 20-step limit. Node and Python wait for the result. The cURL example starts a job and shows how to check it.
1// yarn add @hyperbrowser/sdk2import { Hyperbrowser } from "@hyperbrowser/sdk";34const client = new Hyperbrowser({5 apiKey: process.env.HYPERBROWSER_API_KEY,6});78const result = await client.agents.hyperAgent.startAndWait({9 task: "Open https://news.ycombinator.com and return the title and URL of the top story.",10 maxSteps: 20,11});12console.log("Status:", result.status);13console.log("Result:", result.data?.finalResult);Supported agents
Each agent's documentation lists its available models and configuration options.
| Agent | Provider | Interaction |
|---|---|---|
| HyperAgent | Hyperbrowser, open source | Playwright-based browser automation |
| Browser Use | Open source | DOM-based automation with optional vision |
| Claude Computer Use | Anthropic | Computer-use tools for browser interaction |
| OpenAI CUA | OpenAI | Screenshot-based computer use |
| Gemini Computer Use | Computer-use tools with combined actions | |
| Meta Computer Use | Meta | Computer use with configurable reasoning effort |
| Grok Computer Use | xAI | Computer use with configurable reasoning effort |
| Jev | Jev | DOM-based automation with TypeSafe decision models |
Agent pricing
Browser sessions used by agents are billed at browser rates in addition to steps or tokens.
| Usage | Credits | Price |
|---|---|---|
| HyperAgent | 20 credits / step | $0.02 / step |
| Browser Use | 20 credits / step | $0.02 / step |
| Computer-use models | Per input and output token | View model rates |
Common questions
When should I use an agent instead of Playwright?
Use Playwright to write the exact sequence of page interactions. Use an agent to describe a task and let a model choose the actions. Both run in managed browser sessions.
Read moreCan an agent continue in a browser that is already logged in?
Yes. Pass its session ID when starting the task. Set keepBrowserOpen if another task will follow. To reuse the login in a new session later, save it in a browser profile. Stop retained sessions when finished.
Read moreWhat happens when a task cannot finish?
Check the final status and step history to see where it stopped. The agent may have reached its step limit or failed on a page interaction. Revise the task or its limits before retrying. HyperAgent also has a stop endpoint for cancelling a running job.
Read moreCan I use my own model API keys?
Yes. Set useCustomApiKeys and provide credentials for a provider supported by that agent. Model usage is billed to your provider account, and Hyperbrowser bills browser time. Each agent's documentation lists its supported models and options.
Read moreOther products
Start with a free account.
Includes 5,000 credits for browsers, sandboxes, and web data.