Cloudflare Launches Full AI Agents Platform — Inference, Storage, Voice, and Browsers Unified
Cloudflare announced a comprehensive AI agents platform spanning a unified inference layer for 14+ model providers, Artifacts (Git-compatible versioned storage for agents), Voice Agents, Browser Run with 4x higher concurrency, and Project Think — a next-gen Agents SDK preview. Together they represent Cloudflare's bid to be the infrastructure layer for production AI agents at scale.
Original sourceCloudflare has quietly assembled what may be the most complete infrastructure stack for AI agents outside of the major cloud providers. A cluster of announcements positions it as the place to run agents that need compute, memory, browsers, voice, and storage — all on one global network.
The centerpiece is a unified inference layer that routes to 14+ model providers via a single Workers AI binding, abstracting away provider lock-in. Alongside it, Artifacts provides Git-compatible versioned storage specifically designed for agent workflows — agents can fork, commit, and share state via URLs that any Git client can clone.
Browser Run, already a standout product, got a 4x concurrency bump and added Live View with session recordings and human-in-the-loop interruption. Cloudflare Sandboxes graduated to GA, giving agents persistent isolated environments with background processes. Voice Agents adds a real-time STT/TTS pipeline over WebSockets.
Project Think — a preview of the next Agents SDK generation — promises to move from lightweight primitives to a batteries-included platform where agents can think, act, and persist state across long-running tasks.
The play here is clear: Cloudflare's edge network means low-latency inference globally, and combining that with durable storage, browser control, and voice in a single platform removes the multi-vendor complexity that currently makes production agent deployments painful.
Panel Takes
“Cloudflare just made the multi-vendor agent stack problem disappear. Running inference through Workers AI, persisting state in Artifacts, and controlling browsers through Browser Run in a single Cloudflare Worker with no egress fees between them — that's a genuinely different cost and complexity model than stitching together AWS Lambda + S3 + Browserbase + OpenAI.”
“Cloudflare is adding everything at once and almost none of it is GA. Project Think is a preview, Voice Agents is experimental, Artifacts just launched. 'Batteries-included' platforms have a history of being everything at 80% completion and nothing at 100%. AWS has been promising unified agent infra for two years. The execution risk here is real.”
“The significance of Cloudflare's edge network for agents is underappreciated. Latency isn't just a performance metric — it's the difference between agents that feel responsive and agents that feel broken. A globally distributed agent platform where inference, storage, and browser control are co-located changes what's buildable, especially for real-time voice and interactive agent applications.”