Your agent needs a computer, not a container — introducing @cloudflare/computer
15:15 · August 3, 2026 · Cloudflare AI Blog

Agents need more than just a container to scale. We're introducing @cloudflare/computer, an agent runtime that dynamically orchestrates between fast, efficient isolates and full Linux containers to give every agent a computer of its own.
Summary
Cloudflare has released an early preview of @cloudflare/computer, an open-source package that supplies AI agents with a workspace abstraction resembling a dedicated computer. The runtime supplies a durable virtual filesystem, backed by SQLite and populated from git repositories or storage buckets, together with a consistent set of file, git and shell tools. Execution can be delegated to different backends: lightweight isolates for most file manipulation, data processing and JavaScript tasks, or full Linux containers when native binaries, npm packages or other system-level capabilities are required. The same filesystem remains visible to whichever backend is chosen, so agents can move between environments without explicit data movement.
The design directly addresses the scaling limits of container-per-agent architectures. Rather than provisioning a container for every concurrent agent—an approach the company argues cannot reach hundreds of millions or billions of instances—@cloudflare/computer keeps the agent harness inside a Durable Object and invokes a container only on demand. Frontier models are reported to select the appropriate backend reliably when the tool descriptions indicate the cost and capability trade-offs, keeping container usage below ten percent for many coding and documentation workloads.
All operations through the workspace are gated, audited and observable, giving developers fine-grained control over permitted actions and a persistent record of changes. The package is distributed via npm and exposes both a direct workspace API and an AI-SDK-compatible toolkit, allowing the same primitives to be used both inside an agent loop and for preparatory setup before prompting.
Why it matters
This article is relevant for security professionals as it introduces a new paradigm for sandboxing AI agent execution. The built-in gating, auditing, and isolated environments provide essential mechanisms for securing autonomous AI systems and mitigating risks associated with untrusted code execution.











