Cloudflare ships Agent Cloud

Cloudflare announced integration of OpenAI models into an edge‑oriented 'Agent Cloud' intended to run agents closer to users and systems. (startuphub.ai) The offering presents agents as an edge deployment unit rather than just a chat interface. (startuphub.ai)

Cloudflare said on April 13 it is expanding Agent Cloud so companies can run OpenAI-powered agents on Cloudflare’s network instead of on their own servers. (cloudflare.com) An agent is software that can remember state, call tools, wait for scheduled jobs, and take multi-step actions without a person watching every step. Cloudflare’s Agents documentation says each one runs on a Durable Object, a stateful micro-server with its own database, WebSocket connections, and scheduler. (developers.cloudflare.com) OpenAI said enterprises using Agent Cloud can now deploy agents built with models including GPT-5.4, and can deploy agents built on Codex harness to Cloudflare as well. OpenAI also said the Codex harness is now generally available in Cloudflare Sandboxes and is planned for Workers AI later. (openai.com) Cloudflare is pitching the product as infrastructure, not a chatbot window. Its press release says the goal is to move agents from laptop demos to long-running production workloads across Cloudflare’s global network. (cloudflare.com) That pitch builds on Cloudflare’s existing edge stack. The company’s docs say agents can use Workers AI, OpenAI, Anthropic, or Google Gemini; keep built-in SQL state; expose tools over Model Context Protocol; and wake themselves up later to keep working. (developers.cloudflare.com) Cloudflare has been laying the groundwork for months. In a June 25, 2025 blog post, the company said OpenAI’s Agents Software Development Kit handled reasoning and tool definitions, while Cloudflare’s Agents Software Development Kit supplied identity, persistence, and execution across its network. (blog.cloudflare.com) The technical problem is cost and startup time. Cloudflare said this week that container-based sandboxes can take hundreds of milliseconds to boot and use hundreds of megabytes of memory, which becomes expensive if every worker or customer gets multiple agents. (blog.cloudflare.com) Cloudflare’s answer is lighter-weight execution at the edge. In a March 24 post, the company said its Dynamic Worker Loader can start isolated Workers from code generated at runtime, block outbound internet access, and is now in open beta for paid Workers users. (blog.cloudflare.com) The OpenAI tie-in also fits Cloudflare’s broader model-broker strategy. Cloudflare said in August 2025 that AI Gateway connected customers to more than 350 models across six providers, including OpenAI, with unified billing and routing through one endpoint. (blog.cloudflare.com) OpenAI framed the new arrangement as distribution into Cloudflare’s installed base. The company said more than 1 million business customers use OpenAI directly, Codex has 3 million weekly active users, and its application programming interfaces process more than 15 billion tokens per minute. (openai.com) The immediate test is whether developers treat an agent like a deployable service instead of a chat feature. Cloudflare and OpenAI are both betting that the winning setup is one where the model thinks, the agent keeps state, and the runtime sits close to users and company systems. (cloudflare.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.