Cloudflare expands Agent Cloud
Cloudflare said it is expanding Agent Cloud to support long‑running AI agents and is broadening access to models from OpenAI (including GPT‑5.4) and open‑source options through a single interface. The update is pitched as a portability layer so customers can switch providers with minimal code changes. (itbrief.asia)
An artificial intelligence agent is software that can keep working after you stop typing, calling tools, checking results, and resuming later like a background job. Cloudflare said on April 13 it is expanding its Agent Cloud so those agents can run for days, weeks, or months instead of a single request. (cloudflare.com) Cloudflare’s own documentation says long-running agents can survive restarts, wake on demand, and manage work that lasts far longer than any one web request. The company positions that as a way to move agents from laptop demos to production systems on its network. (developers.cloudflare.com) Cloudflare also said customers can use multiple model providers through one interface, including OpenAI models such as GPT-5.4 and open-source models on Workers Artificial Intelligence. Its Agents product page says memory, scheduling, email, real-time communication, and observability are built in. (cloudflare.com) (workers.cloudflare.com) That setup targets a common problem in artificial intelligence software: the model is only one part of the system, while the rest is state, timing, tool access, and recovery after failures. Cloudflare’s Agents documentation says most artificial intelligence apps are still stateless, meaning they answer once and forget, while real agents need memory, schedules, and coordination. (developers.cloudflare.com) The timing lines up with a broader push by Cloudflare to make its network a home for agent workloads, not just websites and application traffic. On April 14, the company also announced Cloudflare Mesh, a private networking product it said is built for connecting agents, people, and multicloud infrastructure in one secure fabric. (cloudflare.com) On the model side, OpenAI introduced GPT-5.4 on March 5 and said it is designed for professional work, with computer-use capabilities and up to 1 million tokens of context in the application programming interface and Codex. OpenAI’s developer guide says GPT-5.4 is aimed at long-running tasks, multi-step workflows, and tool-heavy workloads. (openai.com) (developers.openai.com) Cloudflare has also been adding larger open-source models to Workers Artificial Intelligence. On March 19, it said Moonshot Artificial Intelligence’s Kimi K2.5 became the first frontier-scale open-source model on Workers Artificial Intelligence, with a 256,000-token context window and support for multi-turn tool calling. (blog.cloudflare.com) A portability pitch sits underneath the product update: if model prices, capabilities, or availability change, developers want to swap providers without rewriting the rest of the application. Cloudflare said the expanded Agent Cloud is meant to let customers switch models with minimal code changes while keeping the surrounding agent infrastructure in place. (cloudflare.com) Cloudflare started pushing this direction in 2025 with an Agents software development kit, a JavaScript framework, and Workers Artificial Intelligence updates for tool calling, structured outputs, and longer context windows. The April 2026 release extends that earlier pitch from building agents to running them as persistent services on Cloudflare’s network. (blog.cloudflare.com)
