Sidegent
Learn to build AI agents by actually building them
137 followers
Learn to build AI agents by actually building them
137 followers
Sidegent is a hands-on learning platform for AI agents. The Sidegent Coder writes and runs real code with you in your browser, so you learn by building, from fundamentals to frameworks (Google ADK, LangGraph, OpenAI). Every lesson auto-graded. Made in Malaysia. Start free.
Sidegent
Hey Product Hunt, I'm Farhan, the maker of Sidegent.
I built this because keeping up with AI agents is exhausting. Everything you need is scattered: a framework doc here, a YouTube video there, a Twitter thread, a half-broken notebook, a Discord message you'll never find again. By the time you've stitched it together, the tooling has already moved on. So most people end up reading about agents instead of building them.
Sidegent puts it in one place and makes you build. It's a hands-on platform where every lesson drops you into a real agent in the browser. You configure it, chat with it, break it, then hit "Check my work" and get auto-graded on what you actually shipped. No passive video watching.
A few things it covers:
- The frameworks people are actually hiring for: LangGraph, Google ADK, the OpenAI Agents SDK, plus the core building blocks like tools, subagents, memory, knowledge bases, and MCP.
- A path from zero to production. Start with fundamentals if you're new, or skip to production patterns if you build for a living.
- A built-in tutor. Stuck on a line in a lesson? Highlight it and ask, and it answers in the context of exactly where you are.
The first module is free, free 3 spin up machine labs, no card needed, so you can see if it clicks before paying anything.
Quick context: Sidegent already has 70+ paying members building agents on it. We started in Malaysia (the platform is fully bilingual, English and Bahasa Melayu), and this launch is about opening it up to builders everywhere. I read every piece of feedback, so tell me what's confusing, what felt too easy or hard, and which course you'd want next. Try it and tell me what breaks.
The Coder writing and running real code with you in the browser and auto-grading it is the part that separates this from another scattered pile of framework docs, since the feedback is on code that actually executed. When I am doing the ADK or LangGraph lessons, does the Coder run in a hosted sandbox with model calls proxied through Sidegent, or can I drop in my own OpenAI/provider keys and hit the real APIs? And does the code I build in a lesson come out as a project I can clone and keep running in my own editor, or does it stay locked to the browser environment once the lesson ends?
Sidegent
@noctis06 Yes it is proxied through sidegent, for now we don't support bringing your own key, as we are more focusing to make it easy for beginner to onboard with 0 setup on their local machine.
For now the project lives fully in the browser, I was just been thinking about this, for more technical ppl and builders, I should build something that you can actually bring it in your own environment but still connected to Sidegent, for example an Agent skills that learn module from Sidegent and implement on local something like that.
Thanks for the question :)
Keeping it zero-setup for onboarding is the right call for beginners. On the bring-your-own-environment idea you mentioned: when that ships, would the code I build in a lesson export as a standalone project I can clone and run without Sidegent, or does the grader/runtime stay coupled so it only runs while connected? And even with the proxy today, do I get to see the raw request/response and token usage per call, or is that abstracted away during lessons?
Congrats on the launch! 🚀
I really like the "learn by building" approach. Reading documentation only gets you so far actually debugging and fixing an agent is where the real learning happens.
I'm curious: as frameworks evolve so quickly, how do you keep lessons and projects up to date without learners running into outdated examples?
Sidegent
@prashant_patil14 Hello Prashant! Thanks, yup things do evolves quickly but on the surface things might look fundamentally different, but it is still the same, for example /goal and /loop its still the same thing just different implementation, I build the course there in such a way that it can adapt to the fast changing scene. And also I think after this, it will be more deep diving into each of the AI Agent harness stuff like RAG, Tooling and so on, more machine labs :D
Grass
Love this @muhammad_farhan24 ! The built-in tutor feature is such a good idea. We ran into something similar building @Grass — keeping persistent VMs alive and reliable is way harder than it looks 😅
Sidegent
@sunnyjoshi wow interesting I should explore Grass for my machine labs
The "learn by actually building" angle really clicks for me. So many AI agent courses are just slides and theory, and you forget everything by Friday. Building something real that you keep is way stickier. Curious how you handle the messy debugging part, like when an agent loops or calls the wrong tool.
Question on the auto-grading: there are loads of valid ways to build an agent, so are you grading the behaviour or one expected solution? Asking because I've hired a lot of engineers, and the thing that'd make this worth it is something a learner can actually show a hiring manager, not just a course completion.