Tencent EdgeOne Makers - Ship AI agents like web apps, in minutes.

@cora998 Couldn't agree more — "why did the agent do that?" is impossible to answer without proper traces. We wanted that visibility built in from the start, not an afterthought. Appreciate you calling it out!

Absolutely. That’s exactly the kind of pain tracing is meant to solve.❤️

@nikita_zaro Python runtimes can have slower cold starts than Node when the framework dependencies are large (CrewAI, for example). Our platform automatically pre-warms the instance when you visit an Agent site, so the first request avoids a cold start.

If you try it out, would love to hear how the first-request latency feels on your side.😊

That's really helpful feedback 😊

Yes, agents can be built and deployed through our CLI. We'll take a look at making that more obvious in the product messaging.

You can find the quick start guide here:

https://pages.edgeone.ai/document/agents-quick-start

@xichiwoo Thanks! Yes — there's a full CLI. npm i -g edgeone, then edgeone makers dev runs your agent locally for debugging, and you deploy straight from the CLI too. Agents are first-class there, same as web apps. :)

@xichiwoo Thanks! And yeah — CLI is fully supported. You can deploy agents straight from the command line with our CLI, fits right into your existing dev workflow. Appreciate the feedback on making that clearer though, we'll make sure it's more visible on the page!

@andreea_stoica4 Your graph definition—nodes, edges, and state—can remain almost entirely unchanged. The required changes are mostly limited to the "glue layer": the entry function and infrastructure integration, which typically takes around 10 lines of code.

✅ No Changes Required (Your Core Business Logic)

StateGraph definitions、Conditional edges、State schemas

Prompt templates、LLM invocation logic

Custom tool implementations

Framework-native configurations such as recursion_limit

These components form the core of your Agent and can be reused as-is.

Required Changes (Around 10 Lines of Code)

Project structure: Move your code to agents//index.py

Entry point: Wrap your logic in an async def handler(context)

function Checkpointer: Use context.store.langgraph_checkpointer

For a step-by-step guide, please refer to: https://pages.edgeone.ai/documen...

That's exactly the experience we were aiming for. The less time spent adapting code to a platform, the more time you can spend improving the agent itself.

@trify3d The sandbox tool built into Makers Agent is underpinned by Tencent Cloud's isolated instances. These instances are specifically designed to host the "side effects" generated during Agent execution—such as running commands, reading/writing files, controlling browsers, and executing code. This enables the LLM to truly possess "hands-on" capabilities, rather than being limited to text output.

Easy misunderstanding. When we say "sandbox," we're talking about an isolated execution environment for agent actions, rather than the browser sandbox most people are familiar with.

@lily_liu8 Thank you! That's exactly the principle behind our design: Agent developers shouldn't have to spend a significant amount of time stitching infrastructure together.

We hope this helps you spend less time assembling infrastructure and more time building what truly matters. If you have any feedback, we'd love to hear it!

@lily_liu8 Appreciate that 😊 We've definitely felt that pain ourselves, which is a big part of why Makers exists in the first place.

@lily_liu8 Right? Stitching five services together just to get one agent live is exhausting — and then you're maintaining all of them. Having it as one coherent stack is the whole point. Glad it feels that way, thanks!

@xiao_zhang9 That’s exactly where things start to matter. A lot of setups work fine in single-user or demo mode, but once you move to real customers, isolation and predictable behavior become essential pretty quickly.😊

@xiao_zhang9 100%. The moment you go from demo to real customers, isolation stops being optional — it's table stakes. We made it a default so you don't have to build that layer yourself. Thanks for calling it out!

@ll_wen Thanks! 🙌 glad you like it

@ll_wen Thanks, really appreciate that! Great team makes all the difference.

@arthur_winston3 Thanks! ❤️

@arthur_winston3 Thank you! 🙏

@hanzhizhang0405 Thanks — that’s a really good way to put it 😊 We’ve seen that the real friction usually starts after the first working version, when you need to actually run it reliably for users, not just build it.

@hanzhizhang0405 That's exactly the gap we saw — tons of tools help you build an agent, very few actually help you run it for real users. We wanted to own that "run and serve" part so you don't have to piece it together yourself. Appreciate you seeing the distinction!