What is a cloud browser? For developers and AI agents

Browsers were built for people to read pages, click links, and fill in forms. As more work moved onto the web, developers needed to automate those interactions, scrape data, and run tasks at a scale that manual browsing couldn’t support.

AI agents bring another source of demand. An agent checking an order or submitting an application needs a browser it can control, often on websites with no suitable API. Running many of these tasks at once takes infrastructure beyond a browser on a laptop.

Cloud browsers provide that infrastructure. They run on remote servers, giving automation scripts and AI agents browser sessions they can control without hosting and maintaining the browsers themselves.

At TinyFish, we build browser infrastructure and web agents that use it to complete tasks. In this guide, we’ll explain how cloud browsers work, what happens inside the provider, and which services are worth considering.

What is a cloud browser?

A cloud browser is a web browser hosted on a remote server. It loads websites and runs their JavaScript. People and applications control it remotely to navigate pages, fill in forms, and retrieve information.

Let's say you're building an agent that checks an order’s delivery date on a supplier’s website. To accomplish that, the agent needs to sign in, find the order, and read its details.

A cloud browser gives it a browser session: a running browser and pages it can control through code.

You could run that browser on your laptop, but if you want your agent working 24/7, you’d have to keep the laptop awake and connected too.

A cloud browser lets your agent use a browser on the provider’s servers instead.

How do cloud browsers work?

Cloud browser providers run browsers on their own machines and keep your application connected to the right one. A session goes through four stages:

1. The provider starts a browser. When your application requests a session, the provider chooses a machine to run the browser. It gives the session an ID and keeps track of which browser belongs to it.
2. Your application connects to that browser. The provider returns a connection address. Your code sends instructions through this connection, such as opening a page or clicking a button. The provider directs those instructions to your session's browser.
3. The browser interacts with the website. It downloads the page and runs its JavaScript on the provider's machine. It can then send page text or screenshots back to your application. There are two connections here: your application talks to the browser, and the browser talks to the website.

TinyFish’s Browser Context Profiles save cookies and browser storage so later agent runs can reuse saved login state.

How are cloud browsers different from headless browsers and AI browser agents?

A headless browser runs without displaying a browser window. It still loads and renders web pages and runs JavaScript, but you just can't see it. Chrome, for example, can run with its usual window or in headless mode. Your code can control it in either mode.

The term cloud browser describes where the browser runs: on a remote cloud server. That browser can run with a graphical window or in headless mode, depending on the service and setup.

For example, headless Chrome running on your laptop is a local headless browser. Headless Chrome running on a cloud server is both a headless browser and a cloud browser.

An AI browser agent controls a browser to complete a task you describe. It uses an AI model to interpret the page and decide what to click or type next.

What should you evaluate in a cloud browser?

• Task success and speed. Running the same agent code on the same websites lets you compare completion rates and time per task.
• Compatibility. Your browser and automation library need to work with the service's remote connection method.
• Login persistence. With saved login state, your agent can start a new session without signing in again.
• Session limits. Limits apply to how many browsers can run at once and how long each stays open. For instance, Browserbase's free plan allows only 3 simultaneous browsers for up to 15 minutes each.
• Debugging. Recordings and screenshots help you trace failures.
• Cost. Well, not much to explain there!

Which cloud browsers are worth considering?

Many cloud browser platforms now offer their own agents too. You can control the browser from your own code, or describe a task and let the provider’s agent work through the website

TinyFish, Browserbase, and Skyvern let you use a cloud browser from your own code or hand a task to their agents. Obviously, we're a bit biased towards our own product, TinyFish, but we'll be fair to our competitors too!

TinyFish

At TinyFish, we started with a web agent and built browser tools to support it.

With the Browser API, your code or your own agent controls a remote Chromium browser, the open-source browser behind Chrome.

TinyFish operates the browser, with residential proxy routing and anti-bot handling included.

With the Agent API, TinyFish's agent decides what to do on the website. You send a URL and a plain-English goal, and it works through the pages and returns structured results. If you're building your own agent, it can ask TinyFish to handle the website task and use the result in its next step.

TinyFish’s Playground browsing Jane Street’s careers site and returning matching roles. Demo source.

For ongoing workflows, TinyFish provides:

• Saved browser profiles that reuse cookies and login state across Agent API runs.
• Parallel runs for tasks across different accounts or websites.
• Logs and screenshots that let you inspect how a run reached its result.

Browser sessions cost $0.002 per minute, including proxy routing. Web Agent costs $0.016 per agent step, so its cost depends on how many steps a task takes.

Browserbase

Browserbase combines hosted Chromium browsers with tools for building and debugging browser agents. You can connect your existing automation code or use Stagehand, its open-source library that mixes normal code with instructions such as “click the submit button.”

Its browser service includes:

• Session recordings and logs for tracing failures.
• A live view where a person can take over the browser.
• Saved browser contexts for reusing login state between sessions.

Browserbase’s Session Inspector combines a recording with session details and playback controls.

With Stagehand, your application still defines the workflow and chooses which steps use AI. Browserbase also offers a hosted Agents API that accepts a task prompt, so you can delegate a complete task instead of writing its workflow.

The Developer plan costs $20 per month for 100 browser hours, then $0.12 per hour. TinyFish charges $12 for 100 browser hours, 40% less than that subscription. Browserbase includes 25 simultaneous browsers; TinyFish starts with 5.

Stagehand itself is free. Model usage is billed separately, and Browserbase proxy traffic over the included 1 GB costs $12 per GB.

Skyvern

Skyvern focuses on repeatable browser workflows, such as downloading documents or filling forms across portals. Its agent reads a screenshot and the page's buttons and fields to choose its next action, then uses Playwright to carry it out.

You can describe a job in Skyvern's app or send it through the API. Its SDK also lets your code control a cloud browser, mixing Playwright instructions with individual AI actions.

Its workflow tools include:

• A visual editor with conditional steps and loops.
• Saved browser profiles for reusing login state.
• Browser recordings and per-step screenshots for inspecting failures.

Skyvern Studio places its browser view beside the workflow editor and chat panel.

Skyvern Cloud operates the browsers for you. You can also self-host the open-source version, which means you run the infrastructure and connect your own AI models.

The Hobby plan costs $29/month for 30,000 credits and up to 10 simultaneous runs. Agent actions and browser time consume credits.

Browser time uses 200 credits per hour, about $0.19 at the Hobby plan's credit price. TinyFish charges $0.12/hour, roughly 38% less for browser time.

How can you get started with TinyFish?

You can run your first TinyFish task with an API key and a single request. Create a key and save it in your terminal.

export TINYFISH_API_KEY="your_api_key_here"

Ask TinyFish’s agent to complete a task

Then send this request to have TinyFish find two product names and prices:

curl -N -X POST https://agent.tinyfish.ai/v1/automation/run-sse \
  -H "X-API-Key: $TINYFISH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://scrapeme.live/shop",
    "goal": "Extract the first 2 product names and prices. Return JSON."
  }'

TinyFish starts the browser and navigates the site for you. Progress updates and the task result return to the same terminal.

Create a browser for your own code

If you want a browser your own code can control, the same key works with the Browser API. This request creates a session and opens the TinyFish homepage:

curl -X POST https://api.browser.tinyfish.ai \
  -H "X-API-Key: $TINYFISH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://www.tinyfish.ai"}'

The response includes cdp_url, the connection address your code uses to control the browser. Playwright connects through chromium.connectOverCDP(cdp_url).

The Playwright example shows how to read the page title and take a screenshot.

When your code finishes, it can end the session using the returned session_id.

You can also try a task in the TinyFish Playground without writing code.