Cloudflare's AI Platform: an inference layer designed for agents

Cloudflare has launched its AI Platform, positioning it as a unified inference layer specifically designed for the complex needs of AI agents. Unlike simple chatbots that make single API calls, agents often chain multiple calls across different models for tasks like classification, planning, and execution. This multi-model reality creates operational chaos: developers are locked into providers, struggle with cascading failures from a single slow API, and lack holistic cost visibility across an average of 3.5 different models per company. Cloudflare's platform tackles this by offering one catalog and one unified endpoint.

Starting today, developers using Cloudflare Workers can call third-party models—from providers like OpenAI, Anthropic, Google, and Alibaba Cloud—using the same `AI.run()` binding used for Cloudflare's own Workers AI models. Switching providers is a one-line code change. The platform provides access to over 70 models spanning text, image, video, and speech, all billable through one set of Cloudflare credits. The integrated AI Gateway offers centralized cost monitoring, automatic retries, and granular logging, allowing teams to track spend by user, customer, or workflow using custom metadata.

Furthermore, Cloudflare is addressing the need for custom models by developing a 'bring your own model' feature, leveraging Replicate's Cog technology to containerize machine learning models for deployment on its infrastructure. This move consolidates the fragmented AI toolchain, giving developers the flexibility to choose the best model for each subtask without vendor lock-in, while ensuring performance and reliability at the edge for latency-sensitive agentic applications.