Cloudflare computer

Overview

Cloudflare computer is an open-source agent runtime that gives AI agents their own operating system on Cloudflare’s edge network. Each agent gets a sandboxed container with a real browser, a real filesystem, and a real shell. This is not a tool-calling abstraction or an API shim — agents interact with a full environment the same way a developer would, but at the scale and latency of Cloudflare’s global infrastructure.

The project launched to immediate traction, hitting #1 on GitHub trending with over 2,690 stars gained in a single day. The interest reflects a gap in the agent infrastructure market: most agent frameworks handle orchestration and prompting, but punt on the runtime environment. Cloudflare computer fills that gap by providing the actual compute substrate agents need.

Architecture

The runtime is built on two Cloudflare primitives: Workers for serverless compute and Durable Objects for stateful coordination. When an agent needs an environment, the platform provisions a container at the nearest edge location, attaches persistent state via Durable Objects, and exposes browser, filesystem, and shell interfaces to the agent process.

Each container is isolated — agents cannot access each other’s environments, and resource limits prevent runaway processes from affecting neighboring workloads. The browser instance supports full page rendering, JavaScript execution, and programmatic interaction, making it suitable for agents that need to navigate websites, fill forms, extract data, or interact with web applications.

Key Capabilities

Edge-native execution: Agents run in whichever of Cloudflare’s 300+ data centers is closest to the user or target system. This matters for latency-sensitive agent workflows — a web-scraping agent targeting European sites runs from European edge nodes, not a US-East data center.

Browser-first agent runtime: The built-in browser is not a headless afterthought. It is a core primitive designed for agents that need to interact with the web — login to services, navigate multi-step workflows, capture screenshots, and extract structured data from rendered pages.

State persistence: Durable Objects provide built-in state management. Agents can checkpoint their progress, survive container restarts, and resume long-running tasks without external state stores. This is infrastructure-level resumability, not application-level retry logic.

Native Cloudflare integration: Agents running on the platform can access Cloudflare’s broader service catalog — R2 for object storage, D1 for SQLite databases, Workers AI for inference, and KV for key-value state — without network hops or authentication ceremony.

Use Cases

Web automation agents that need to interact with sites at scale. Data extraction pipelines that require browser rendering to access JavaScript-heavy pages. QA and testing agents that validate web applications across geographies. Multi-step workflow agents that need persistent state and reliable execution. Any agent workload where you want the infrastructure managed and globally distributed rather than self-hosted.

Considerations

The platform runs on Cloudflare’s infrastructure, which means your agent workloads are subject to Cloudflare’s pricing, usage limits, and terms of service for Workers and Durable Objects. While the runtime itself is open source, production deployments require a Cloudflare account. Teams evaluating this should compare the economics against self-hosted container solutions (Docker, Kubernetes) for their specific workload patterns — edge distribution is a clear win for geographically distributed tasks but may not justify the overhead for centralized batch processing.

The project is early. Documentation and community patterns are still developing. Expect rough edges and breaking changes as the API surface stabilizes.

Who It’s For

Platform teams building agent infrastructure who want managed, globally distributed compute. Developers building web-interacting agents who need reliable browser automation at scale. Companies already in the Cloudflare ecosystem who want to extend their edge infrastructure to agent workloads. Open-source contributors interested in the intersection of edge computing and AI agents.