AI web-reading tools
Give agents a short-lived browser for sites that require JavaScript rendering, DOM access, or screenshots but do not need full Chromium fidelity.
Independent tool overview
Kitesurf is Cloudflare's beta browser engine for AI agents and one-shot web automation. It runs in V8 isolates on Cloudflare Workers, exposes a subset of the Chrome DevTools Protocol, and can be selected in Browser Run Quick Actions or CDP sessions. Compared with warm Chromium in Cloudflare's 14-URL test, Kitesurf used materially less CPU and memory for screenshots and HTML extraction but took longer in wall-clock time. It is a useful fit for ephemeral extraction, screenshots, PDFs, and compatible browsing tasks—not a full Chromium replacement for persistent logins, video, WebGL, bot challenges, or pixel-perfect rendering.
Visit the official Kitesurf site ↗
Overview
Cloudflare introduced Kitesurf on August 6, 2026 as an agent-first alternative to running a full human-oriented browser. It removes capabilities that an automated model often does not need and focuses on structured page access, DOM inspection, screenshots, extraction, scalability, isolation, and lower compute consumption.
Kitesurf runs entirely on Cloudflare Workers. Its Engine exposes HTTP and CDP endpoints and holds the short-lived session state; separate PageScript and PageRenderer components process untrusted page code and visuals. A dedicated outbound component controls network access, applies browser-shaped headers and CORS policy, filters responses, and keeps cookies isolated by page.
Developers select Kitesurf by adding browser=kitesurf to a Browser Run Quick Action or CDP endpoint. Existing Puppeteer, Playwright, chrome-remote-interface, and CDP-compatible MCP clients can connect, but Kitesurf implements only the protocol surface needed for current agent and automation scenarios.
The engine is intentionally ephemeral and stateless wherever possible. That makes it well suited to bursty one-shot tasks and recovery by replay, but it is the wrong choice when a workflow needs a long authenticated session or cross-request browser state.
Cloudflare reports that Kitesurf passed more than 215,000 Web Platform Tests at launch. Its own five-run, 14-URL comparison found 3.1–7.0 times lower CPU or memory for screenshot and extraction work, while warm Chromium completed the same tasks about 1.7–1.8 times faster in wall time. Those are provider-run medians on a small corpus, so teams should benchmark their own sites.
Kitesurf starts every page from untrusted input, but isolation does not make an AI browsing workflow safe by itself. Agents can still follow malicious page instructions, leak authorized data through permitted actions, submit the wrong form, or violate a site's rules. Use domain and action restrictions, short-lived credentials, human approval, audit logs, and a Chromium fallback for unsupported pages.
Use cases
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Give agents a short-lived browser for sites that require JavaScript rendering, DOM access, or screenshots but do not need full Chromium fidelity.
Generate a screenshot, PDF, rendered document, link list, snapshot, or structured extraction from a compatible URL.
Start isolated sessions only for the duration of a task and scale many independent jobs without maintaining a warm browser fleet.
Connect existing Puppeteer, Playwright, chrome-remote-interface, or MCP tooling through a Cloudflare endpoint with a browser selector.
Reduce CPU and memory for compatible extraction and screenshot tasks when those resources matter more than minimum wall-clock latency.
Route simple pages to Kitesurf and complex or stateful pages to Chromium after testing both against explicit correctness and rendering criteria.
Capabilities
Runs the engine, page scripting, rendering, and controlled outbound access on Cloudflare's serverless platform rather than launching desktop Chrome.
Supports Browser Run's one-request endpoints by adding the browser=kitesurf query parameter for operations such as screenshots and extraction.
Exposes a Chrome DevTools Protocol WebSocket so compatible automation libraries and agent clients can drive a short-lived page.
Lets developers enter a URL, inspect the rendered page, explore DOM and console output, watch network activity, and view isolate memory without first writing code.
Treats every page as untrusted and starts each session clean to reduce state leakage and make failures recoverable by replay.
Routes page asset and fetch requests through a dedicated outbound worker that enforces policy, response filtering, headers, CORS, and separate cookie jars.
Implements the HTML, DOM, selection, SVG, XHR, CSS, and JavaScript behavior needed for many agent-oriented web tasks.
Builds page visuals without a full Chromium stack, prioritizing compute efficiency over perfect visual compatibility.
Cloudflare reports successful rendering for TodoMVC variants across React, Vue, Angular, Preact, and vanilla JavaScript plus several content and dashboard sites.
Uses a growing WPT suite and continuous performance tests to expand standards coverage and catch regressions.
Lives inside Browser Run, allowing developers to choose standard Chromium when a page needs capabilities Kitesurf does not yet implement.
Process
Step 1
Use ordinary HTTP fetch for static data, Kitesurf for compatible rendered one-shot work, and Chromium for video, WebGL, difficult bot challenges, pixel-perfect output, or persistent login state.
Step 2
Use a Browser Run API token with the minimum account permissions, store it in a secret manager, rotate it, and never expose it to page JavaScript or model output.
Step 3
Test the exact target page, inspect DOM, network, console, memory, screenshot fidelity, and required interactions before committing it to automation.
Step 4
Record required text, fields, links, screenshots, actions, and error behavior for representative pages, languages, frameworks, cookie states, and redirects.
Step 5
Add browser=kitesurf to the Quick Action or CDP URL; do not assume a default Browser Run session uses this engine.
Step 6
Allow only required schemes, domains, redirects, ports, and download types; block local, metadata, administrative, and sensitive internal endpoints.
Step 7
Keep webpage text and scripts in the data boundary, ignore instructions that attempt to change agent policy, and prevent pages from requesting credentials or broader tools.
Step 8
Prefer read-only extraction; require explicit confirmation for logins, submissions, purchases, uploads, messages, deletions, or any action affecting another person or system.
Step 9
Cap session time, redirects, pages, retries, concurrency, output size, and total browser hours; record X-Browser-Ms-Used where applicable.
Step 10
Compare extracted values against the DOM or authoritative source, inspect screenshots and PDFs, and reject partial pages, missing controls, or silently blocked requests.
Step 11
Detect unsupported APIs, low-fidelity rendering, login persistence, challenge pages, and timeouts, then retry through Browser Run's Chromium engine when allowed.
Step 12
Track completion, accuracy, latency, CPU, memory, browser time, rate limits, site-level failure, policy violations, and engine-version changes.
Cost
Cloudflare states that Kitesurf is free while in beta and subject to per-account limits. Browser Run itself is available on Workers Free and Paid plans with published browser-hour and concurrency pricing for its general service. Because beta treatment may change, confirm the dashboard and current Kitesurf documentation before forecasting production cost.
Free during beta
Current Kitesurf engine access through Browser Run, behind per-account limits.
10 browser minutes per day included
The standard Browser Run free-plan allowance and limits surrounding the service.
10 hours/month included, then $0.09/hour
The standard paid Browser Run model; verify how it will apply to Kitesurf during and after beta.
Pricing checked . Check current pricing at the source ↗
Assessment
Compare
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Data Analysis
A web-automation and structured-query platform for teams that prefer high-level semantic extraction over direct browser-engine control.
Explore AgentQL →Agents
A specialized computer-use model for visually operating browser interfaces when model-driven UI actions matter more than a lightweight engine.
Explore Gemini Computer Use →Agents
A managed end-user agent with its own browser and tools for people who want delegated tasks rather than developer infrastructure.
Explore ChatGPT Agent →Consumer
A user-facing AI browser experience for interactive browsing, unlike Kitesurf's ephemeral headless developer runtime.
Explore ChatGPT Atlas →Questions
Kitesurf is Cloudflare's lightweight, stateless browser engine for AI agents. It runs in V8 isolates on Workers and is accessed through Cloudflare Browser Run Quick Actions or a CDP endpoint.
No. It is a headless engine built for automated tasks. It intentionally omits human-oriented features such as tabs, extensions, synchronization, and pixel-perfect interactive rendering.
Cloudflare says Kitesurf is free while in beta, behind per-account limits. The surrounding Browser Run service has Workers Free and Paid allowances, so recheck the current dashboard and documentation before production budgeting.
Add browser=kitesurf to a Browser Run Quick Action or CDP endpoint. You need a Cloudflare account ID and scoped Browser Run token for API access; the public playground offers a faster compatibility test.
It exposes CDP for Playwright, Puppeteer, chrome-remote-interface, and CDP-based MCP clients. Kitesurf implements only part of CDP, so test the exact methods, navigation, events, and page features your automation needs.
It depends on the metric. In Cloudflare's small launch benchmark, Kitesurf used 3.1–7 times less CPU or memory for screenshots and HTML extraction, while warm Chromium finished about 1.7–1.8 times faster in wall time.
Short in-session interactions may work on compatible pages, but Cloudflare specifically says long-running authenticated sessions requiring persistent state are not a good fit. Protect credentials and use Chromium when continuity is required.
Not as of this review. Use Browser Run's default Chromium engine for video, WebGL, high-fidelity graphics, and other unsupported browser APIs.
It is not designed to negotiate challenges that require genuine browser TLS fingerprints. Do not use browser automation to evade access controls; use authorized APIs or Chromium only where the site's rules permit automation.
No browser engine can solve the full agent problem. Kitesurf isolates sessions and network components, but malicious page content can still manipulate the model. Keep webpage instructions untrusted, restrict tools and destinations, and require approval for consequential actions.
Choose it for compatible, ephemeral extraction, screenshots, PDFs, and agent browsing when CPU and memory efficiency matter. Choose Chromium for persistent sessions, higher fidelity, video, WebGL, difficult challenges, or broader CDP coverage.
Cloudflare labels it beta and says it is continuing CDP, rendering, WPT, and efficiency work. A controlled production pilot may be reasonable if you test target sites, validate outputs, implement Chromium fallback, and monitor regressions.
Bottom line
Kitesurf is a compelling beta for developers whose agents need brief, isolated access to rendered web pages and can tolerate lower browser fidelity. Its value is compute efficiency and serverless scale, not raw wall-clock speed or universal compatibility. Start with the playground, benchmark the exact URLs and actions that matter, lock down destinations and credentials, validate every extracted result, and keep a Chromium fallback. Avoid it for persistent authenticated work, video, WebGL, challenge-heavy sites, or any task where a subtle rendering difference can cause a consequential error.
Visit Kitesurf website ↗
Inkling-Small - Thinking Machines' compact open model that rivals full-size version

Ramp’s endpoint that matches each AI request to the cheapest fitting model
.jpeg)
Lyria 3.5 - Google's new music model with more realistic vocals

Google's video model update with scene extensions and 4K upscaling

Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.