agents: what it is, what problem it solves & why it's gaining traction

What it solves

Cloudflare Agents provides a framework for building and deploying persistent, stateful execution environments for agentic workloads. It solves the problem of managing the lifecycle, state, and communication of millions of individual agents (e.g., one per user or session) that can hibernate when idle and wake on demand, reducing costs and operational overhead.

How it works

Built on Cloudflare Durable Objects, each agent is a stateful entity with its own storage and lifecycle. The framework provides a type-safe RPC system via the @callable() decorator, allowing clients to call methods on agents as if they were local functions. It supports real-time synchronization of state changes to connected clients via WebSockets and integrates with AI models, MCP (Model Context Protocol), and durable multi-step workflows.

Who it’s for

Developers building AI agents, real-time collaborative applications, or stateful services that require persistent state and AI integration on the edge.

Highlights

• Persistent State: State survives restarts and syncs automatically to all connected clients.
• AI Capabilities: Includes specialized packages for AI chat, voice pipelines (STT/TTS), and "Code Mode" where LLMs generate executable TypeScript code.
• Sandboxed Execution: Ability to run LLM-generated code in an isolated Worker with a virtual filesystem.
• MCP Support: Can act as MCP servers or clients to extend agent capabilities.
• Infrastructure Integration: Built-in support for scheduling, email, and x402 pay-per-call payments.
• Frontend Integration: Dedicated React hooks (useAgent, useAgentChat) for seamless frontend connectivity.