Native Support for AI Agents
Chris ChenAI Agents have become a baseline requirement for developers, yet building and deploying them still carries a non-trivial barrier — there is too much infrastructure you have to wire up yourself. EdgeOne Makers (the rebrand of EdgeOne Pages) adds native Agent support on top of its existing full-stack web capabilities, making Agent development as simple as web development. Developers focus on business logic; the platform handles the underlying infrastructure. Web and Agent code can be deployed in the same project, sharing the same domains, configuration, deployment pipeline, and observability stack.
What This Upgrade Brings
Agent Runtime. An Agent Runtime is provided that supports running mainstream frameworks such as Claude Agent SDK, OpenAI Agents, LangGraph, DeepAgents, and CrewAI directly. Handlers stay in each framework's native idiom — the platform does not introduce a new SDK abstraction.
Session Storage. context.store exposes each framework's native session-storage type — for example, OpenAI Agents gets Session, LangGraph gets BaseCheckpointSaver. By default frameworks only ship abstract protocols, and persistence usually requires you to wire up Postgres or other cloud storage. Makers backs all of them with the platform's Blob storage out of the box, with no extra services to enable.
Tools and Sandboxes. Common tools such as files_* and browser_* are pre-packaged in each framework's native protocol — drop them into the tools array and they just work. The underlying sandbox is managed per conversation lifecycle: a micro-VM is lazily started on first use, reused within the conversation, and recycled on timeout. Custom tools can be freely composed via the atomic interfaces on context.sandbox.
Model Gateway. No need to apply for separate API keys — mainstream models are callable out of the box. A limited free model-token quota is provided to help you get up and running quickly; you can also swap in your own keys.
Real-time Observability. edgeone makers dev brings up a local debug panel. With no extra instrumentation, every request's LLM calls, tool calls, and session reads/writes are automatically chained and inspectable. Local and production share the exact same observability primitives.
Getting Started
Run edgeone makers create to pull down a template, then edgeone makers dev locally to start developing. Templates cover the major frameworks as well as minimal framework-free examples in Python and JavaScript. You can also one-click deploy any template directly from the template list to your own account.
For deeper details on usage and design rationale, see A Production-Grade Agent in a Few Lines of Code.