Cloudflare has rebuilt Browser Run on its Containers platform to support up to 120 concurrent headless browser instances — four times the previous limit of 30 — and says the change yields roughly 50% faster response times for quick actions. The upgrade also adds WebGL and WebMCP support and requires no changes from existing users. That combination targets AI agent workloads that demand bursty, low-latency browsing rather than long-lived human sessions.
To handle spiky agent traffic, Cloudflare introduced dedicated container pools with regionally pre-warmed browsers and revised state management. The backend moved from eventual — consistency Workers KV to D1 with Queues, implementing transactional assignment and batch writes tuned to manage up to 500,000 containers per location. For the quick — action path, the team replaced a multi — step WebSocket choreography with single HTTP requests executed entirely inside the container, simplifying the flow and cutting latency.
Cloudflare separates compute and orchestration into two primitives. Dynamic Workers provide V8 isolate — based execution that boots in milliseconds for lightweight tasks such as linting, typechecks, and API calls. Sandboxes, now generally available, supply full Linux containers for git, bash, dev servers, and multi — language builds, and inject credentials securely via an egress proxy so agents never see raw tokens. Orchestration is handled by Dynamic Workflows, a roughly 300 — line, MIT-licensed library that extends Cloudflare’s durable execution engine. The library lets workflow code vary by tenant, agent, or request, and is designed to coordinate the new compute primitives across short — lived agent tasks and containerized browsing sessions.
Agent Memory, currently in private beta, extracts structured memories using a dual-pass ingestion pipeline and retrieves them with a five-channel parallel search that applies Reciprocal Rank Fusion. Shared memory profiles allow teams of agents to access a common knowledge base, supporting coordinated multi — agent behavior without exposing raw context to individual processes.
Browsing is exposed through Browser Run on Containers as controllable headless Chromium instances accessible via the DevTools Protocol or the Agents SDK. The WebMCP addition enables Model Context Protocol interactions directly through the browser. Cloudflare also introduced a commerce protocol co-designed with Stripe that lets agents create accounts, register domains, start subscriptions, and deploy; Stripe handles identity and payments, and the default provider spending cap is $100 per month.
Cloudflare framed the effort as a distinct approach from hyperscaler alternatives. AWS’s Bedrock AgentCore includes an Agent Registry but lacks a managed browser service and no direct built — in equivalent to Agent Memory, while Google Cloud’s GKE Agent Sandbox is a Kubernetes — native primitive rather than a managed platform offering. According to Cloudflare’s account, neither competitor currently combines a managed browsing layer with an integrated commerce protocol in the same way.
Sources
Replies (0)
No replies in this topic yet.
