Tag: Cloudflare

  • Cloudflare Kitesurf: A New Cloud Browser Built for AI Agents

    Cloudflare Kitesurf: A New Cloud Browser Built for AI Agents

    Cloudflare has introduced Kitesurf, a cloud-hosted browser specifically designed to help AI agents navigate websites and complete online tasks. Currently available as a free beta through Browser Run, Kitesurf operates on Cloudflare Workers and is designed to be more resource-efficient than traditional browsers like Chromium.

    What Is Kitesurf?

    Kitesurf is not a consumer browser with a desktop interface. Instead, it is a developer-focused tool accessed through APIs and automation libraries. It allows AI agents to take screenshots, extract HTML, fill out forms, and perform other automated tasks. It supports the Chrome DevTools Protocol, making it compatible with popular automation tools such as Playwright, Puppeteer, and various AI clients.

    This setup eliminates the need for developers to maintain separate browser services or dedicated server infrastructure for each agent task. By running on Cloudflare’s edge network, Kitesurf aims to simplify and accelerate AI-driven web interactions.

    Performance and Efficiency Gains

    Cloudflare’s benchmarks compared Kitesurf with Chromium across five median Browser Run tests on 14 different websites. The results show significant reductions in CPU and memory usage:

    • Screenshots: Kitesurf used 380 milliseconds of CPU time versus Chromium’s 1,173 milliseconds, and 57.8 MiB of memory compared to Chromium’s 271 MiB.
    • HTML extraction: Kitesurf used 229 milliseconds of CPU time versus Chromium’s 877 milliseconds, and 39.4 MiB of memory compared to 273.7 MiB.

    However, these efficiency gains come with a trade-off: Kitesurf took 1.8 times longer for screenshots and 1.7 times longer for HTML extraction. Despite this, Cloudflare describes Kitesurf as “significantly more efficient” due to the substantial resource savings.

    Security-First Architecture

    Security is a core focus of Kitesurf. Each browser session starts fresh, and separate components are isolated to limit data leaks between pages. A dedicated SandboxOutbound worker handles requests for images, scripts, fonts, and stylesheets. It enforces cross-origin rules, filters responses, adds browser headers, and keeps cookies in separate jars. Requests that fail its rules receive a 403 response.

    Kitesurf divides browser work among three components: the Engine (which stores session data and handles public connections), PageScript (which processes HTML, CSS, JavaScript, and WebAssembly), and PageRenderer (which creates screenshots and PDF output). This modular design enhances both security and efficiency.

    Current Capabilities and Limitations

    During its beta phase, Kitesurf can render popular sites like Wikipedia, Hacker News, Cloudflare’s blog, and TodoMVC applications. It passes more than 235,000 Web Platform Test subtests, indicating strong standards support. However, it does not yet support video playback, WebGL, long authenticated sessions, or some bot-check systems. Developers requiring these features can still use Browser Run’s Chromium option.

    Future Plans

    Cloudflare built Kitesurf in just 12 weeks and plans to continue development. Future updates will include broader protocol coverage, better screenshot capabilities, more standards support, and even lower resource usage. The company also intends to release Kitesurf as open-source software once the project matures.

    Kitesurf represents a significant step toward more efficient and secure AI agent operations, offering developers a new tool to streamline automated web tasks while reducing infrastructure overhead.

  • Cloudflare Launches Identity-Aware AI Gateway for Real-Time Employee AI Usage Visibility and Cost Control

    Cloudflare Launches Identity-Aware AI Gateway for Real-Time Employee AI Usage Visibility and Cost Control

    Cloudflare, Inc. (NYSE: NET) has introduced a new tool called Identity-Aware AI Gateway, giving IT and security teams unprecedented visibility into how employees and automated agents use AI models at work. The feature, announced on August 6, 2026, allows companies to see exactly which user or system sent a request to an AI model, set individual or team-level spending limits, and automatically flag risky content—all without requiring new infrastructure or disrupting workflows.

    Accompanying the gateway is a User Insights feature that learns normal AI usage patterns for each person and automated system, alerting teams when behavior deviates from the baseline. This addresses two major challenges: rising AI costs and security blind spots. As AI requests grow cheaper but more numerous, bills can spike unexpectedly, and sensitive data like employee names, passwords, or confidential files may leak to external AI providers unnoticed.

    “When security is clunky, it becomes a speed bump that slows down innovation. Right now, companies are wary of surprise AI bills or accidental data leaks, so their instinct is to lock down access and restrict what teams can build,” said Dane Knecht, CTO of Cloudflare. “Connecting our access controls directly to AI Gateway flips that dynamic. Teams get the freedom to work with any AI model they choose. IT gets real-time visibility, guardrails, and the budget controls they actually need.”

    Key capabilities of the Identity-Aware AI Gateway and User Insights include:

    • See Exactly Who’s Sending What: By integrating with Cloudflare Access, every AI request is tied to a verified identity—whether employee or automated system.
    • Catch Unusual Behavior Early: User Insights builds a baseline of each user’s AI activity and alerts IT when spending spikes abnormally.
    • Identify Efficiency Opportunities: The system flags when employees use high-powered models for simple tasks that cheaper models could handle.
    • Enforce Identity-Based Access: Replace shared API keys with integration into existing Identity Providers (IdP) and Zero Trust (ZTNA) infrastructure, validating user and device posture before sending requests.
    • Unified AI Traffic Control: All requests flow through a single point, enabling caching to cut redundant costs, rate limiting to prevent runaway usage, and automatic filters that strip sensitive data before it reaches external AI providers.

    Max Baumgarten, Security Engineer at Flexport, commented: “Shared API keys make it almost impossible to tell who is using an AI service or apply the access rules we already have for employees. Putting Cloudflare Access in front of AI Gateway gives each request an authenticated identity and lets us use our existing identity policies at the gateway. Our teams can adopt AI tools without creating a separate authentication system for every client.”

    For more details, visit the Cloudflare AI Security blog or the official Cloudflare website.

  • Cloudflare Unveils Wallets and Digital IDs for AI Agents to Enable Secure Autonomous Commerce

    Cloudflare Unveils Wallets and Digital IDs for AI Agents to Enable Secure Autonomous Commerce

    Cloudflare, Inc. (NYSE: NET), the leading connectivity cloud company, today announced Cloudflare Wallets and cloudflare.pay, a new suite of tools designed to give AI agents a stable identity and the ability to make purchases online safely within limits set by their human creators. This marks a significant step toward enabling agentic commerce—where software programs autonomously browse, query APIs, and transact on behalf of individuals and businesses.

    As AI agents become a normal part of doing business online, the Internet—built for humans—lacks reliable mechanisms to verify whether an agent is a legitimate assistant or a malicious actor. Traditional bot detection tools, designed for search crawlers, are inadequate for agents that conduct transactions. Cloudflare’s solution addresses this by providing a verifiable link between each agent and its human or organizational owner, establishing trust and accountability at every step.

    “The Internet is shifting from human-driven browsing to agent-driven commerce, and the infrastructure needs to keep up,” said Matthew Prince, co-founder and CEO of Cloudflare. “When an agent shows up at your door, you need to know who sent it. Cloudflare can give agents a face—a link to the human or organization that owns them—so that trust, accountability, and real commerce can follow. It’s the identity and payment infrastructure the agentic web needs to function.”

    Cloudflare Wallets and cloudflare.pay work together to solve the identity and payment challenge. Each Cloudflare account will receive a unique web address serving as a stable ID, which users can extend to specific agents. This allows any receiving business to see exactly who authorized the agent. Additionally, a Cloudflare “Account Wallet” acts as a central balance for receiving, holding, and managing stablecoins. Users can then assign “Virtual Wallets” to individual agents, each with built-in guardrails such as spending caps, approved merchant lists, and maximum transaction sizes. Combined with the Monetization Gateway, these tools complete the two-sided agentic payment market.

    Cloudflare Wallet handle reservation opens today. Full wallet access, including onramping and offramping funds and the ability to issue Virtual Wallets, will be available in the coming months. Users can claim a handle at cloudflare.pay and learn more on the Cloudflare blog.