Picture by Editor
# Introduction
If you wish to study agent engineering by doing as a substitute of simply studying about it, one of the simplest ways remains to be to fork actual repos, run them regionally, and alter them on your personal use. That is the place the true studying occurs. I’ve hand-picked the perfect 10 ones, the tasks which might be each helpful and well known, so you possibly can see how agent apps are being constructed at this time. So, let’s get began.
# 1. OpenClaw
OpenClaw (~343k ⭐) is the one I might level to first if you wish to see what the subsequent wave of private AI assistants would possibly appear to be. It’s constructed as a private assistant that runs by yourself units and connects to the instruments folks already use, like WhatsApp, Telegram, Slack, Discord, Sign, and iMessage. What makes it fascinating is that it isn’t only a easy chat demo. It looks like an actual assistant product, with multi-channel assist, voice options, and a broader ecosystem round abilities and management. If you’d like a repo that feels near an actual agent system, this can be a robust place to start out.
# 2. OpenHands
OpenHands (~70k ⭐) is a good repo to fork in case your predominant curiosity is coding brokers. It’s constructed round AI-driven growth and now has a wider ecosystem round it, together with cloud, docs, CLI, SDK, benchmarking, and integrations. That issues as a result of you aren’t simply taking a look at one demo. You possibly can research the core agent, verify the interface, and in addition see how the group thinks about analysis and deployment. If you wish to construct or customise a coding assistant, this is among the most sensible repos to study from.
# 3. browser-use
browser-use (~85k ⭐) is among the most helpful tasks if you’d like brokers that may really do issues on the internet. The thought is straightforward: it makes web sites simpler for AI brokers to make use of, to allow them to deal with browser-based duties with much less friction. That makes it simple to experiment with, since plenty of actual agent work leads to the browser anyway — kind filling, analysis, navigation, and repetitive on-line duties. It additionally has supporting repos and examples, which makes it simpler to go from curiosity to one thing you possibly can take a look at in an actual workflow.
# 4. DeerFlow
DeerFlow (~55k ⭐) is among the extra fascinating tasks if you wish to perceive long-horizon agent programs. It’s an open-source super-agent harness that brings collectively sub-agents, reminiscence, sandboxes, abilities, and instruments to analysis, code, and create throughout longer duties. So, it isn’t simply wrapping software calls. It’s making an attempt to handle the complete construction round extra advanced agent conduct. If you wish to see how fashionable agent programs are being constructed round reminiscence, coordination, and extensibility, this can be a very helpful repo to fork.
# 5. CrewAI
CrewAI (~48k ⭐) remains to be one of many best repos to know if you’d like multi-agent orchestration with out an excessive amount of complexity. It’s a quick, versatile framework for multi-agent automation, and it’s constructed independently fairly than on prime of LangChain. The psychological mannequin is straightforward, the setup is approachable, and the docs and examples are pleasant sufficient for inexperienced persons. If you’d like a Python-first repo which you could fork and switch into one thing helpful, CrewAI nonetheless deserves a spot close to the highest.
# 6. LangGraph
LangGraph (~28k ⭐) is the repo to check if you wish to perceive the engineering aspect of brokers, not simply the flashy demo aspect. LangChain describes it as a low-level orchestration framework for long-running, stateful, controllable brokers. It pushes you to suppose when it comes to graphs, state, management circulate, and resilience. It’s particularly helpful if you wish to transfer past easy prompt-plus-tool-call programs and perceive how extra critical agent runtimes are put collectively. It might not really feel as fast to choose up as another repos, however it teaches lots.
# 7. OpenAI Brokers SDK
The OpenAI Brokers SDK (~20k ⭐) is an efficient choice if you’d like one thing light-weight however nonetheless fashionable. It’s constructed as a compact framework for multi-agent workflows, and the docs current it as a production-ready path with a small set of helpful constructing blocks. You get instruments, handoffs, periods, tracing, and realtime patterns with out having to wade by way of an enormous framework. For those who like easy surfaces and direct management, this is among the higher starter repos to discover.
# 8. AutoGen
AutoGen (~56k ⭐) remains to be one of the necessary repos within the multi-agent house. Microsoft frames it as a programming framework for agentic AI, and the docs go additional into enterprise workflows, analysis collaboration, and distributed multi-agent functions. It belongs in this sort of record as a result of there’s a lot to study from it. The orchestration concepts, agent dialog patterns, and framework design are all price learning. It might not be the best place to begin for everybody, however it’s nonetheless one of the influential tasks within the class.
# 9. GPT Researcher (~26k ⭐)
GPT Researcher is a good selection if you wish to research a deep-research agent as a substitute of a normal framework. It’s an autonomous agent for deep analysis utilizing any massive language mannequin (LLM) supplier, and its surrounding materials reveals the way it handles multi-agent analysis and report era. This offers you one clear workflow to check from begin to end. You possibly can see planning, searching, supply gathering, synthesis, and reporting multi function place. If you’d like one thing concrete fairly than summary, this is among the most forkable repos on the record.
# 10. Letta
Letta (~22k ⭐) stands out as a result of it places reminiscence and state on the middle of the agent design. The repo describes it as a platform for constructing stateful brokers with superior reminiscence that may study and enhance over time. This is a crucial angle as a result of plenty of agent repos focus totally on orchestration. Letta widens the image. It’s a good repo to discover if you’d like brokers that persist, keep in mind, and evolve as a substitute of beginning recent each time. For memory-focused agent work, it is among the extra fascinating tasks to fork at this time.
# Wrapping Up
All ten are price cloning, however they educate various things when you really run them and begin altering the code. That’s the place the true studying begins.
Kanwal Mehreen is a machine studying engineer and a technical author with a profound ardour for knowledge science and the intersection of AI with medication. She co-authored the book “Maximizing Productivity with ChatGPT”. As a Google Era Scholar 2022 for APAC, she champions range and educational excellence. She’s additionally acknowledged as a Teradata Range in Tech Scholar, Mitacs Globalink Analysis Scholar, and Harvard WeCode Scholar. Kanwal is an ardent advocate for change, having based FEMCodes to empower girls in STEM fields.
