Product Thumbnail

SWE-Kit

Build your own Devin like software engineering agent

Open Source
Software Engineering
Artificial Intelligence

SWE-Kit is an headless IDE with AI-Native Tools for building custom coding agents with any Agentic Framework & LLMs of your choice.

Top comment

Hey, PH community, I have built SWE-Kit, an open-source LLM toolkit (Function Calling) which makes building state-of-the-art agents specialised in coding like Devin very easy. After building multiple coding agents, I noticed a typical pattern: creating & perfecting tools and integrations was the most time-consuming. So, I built SWE-Kit, which allows developers the flexibility to build reliable coding agents customized for their use cases. You might wonder, aren’t there open coding agents already? After all, many things are going on in this space. Well, yes, but the thing is, most of these are 1. too unreliable for any serious implementations 2. too complicated to customise. The idea is to provide a toolkit so developers of any skill level can build Devin-like software engineering agents with a few lines of code. SWE-Kit’s primary features include: - Optimized Coding Tools: Includes Code Analysis, File Operations, and Shell tools for seamless interaction with codebases and operating systems. - Browser Interaction Tool: Enables navigation and interaction with UI-based applications and codebases. - Framework Agnostic: Compatible with frameworks like LangChain, LlamaIndex, CrewAi, and Autogen, this allows you to work with your preferred setup. - Third-Party Integrations: Connects with applications like GitHub, Slack, Jira, and Gmail to build fully autonomous, end-to-end AI coding agents. - Flexible Deployment: Run on any local Docker host or remote server for adaptable deployment options. To better showcase the SWE kit's capability, we tested it on swebench.com, a comprehensive benchmark for testing coding agents. It scored 48.60%, whereas Devin scored only 13.86%. To help you get started quickly, we’ve created several example agents: - GitHub PR Agent: Autonomously reviews GitHub pull requests with full codebase context. - SWE Agent: Writes new features, debugs code, refactors, and creates tests. - Codebase Q&A Agent: Enables natural language interactions with the codebase. I’d like feedback on how this tool fits into your automation workflows. Please try it out and let me know how it performs.

Comment highlights

@karan_vaidya2 congrats! Wondering how the browser interaction tool works. Do we need to provide Anthropic's recent release of browser controller? Also, how do you handle systems that require authentication, say Linkedin?

Congratulations @karan_vaidya2 and team on launching SWE-Kit! This headless IDE with AI-native tools sounds perfect for building custom coding agents.

Those are some sick stats being thrown around in the video! Congrats on making this, be proud!

The flexibility to choose any agentic framework and LLM makes it so adaptable. Excited to see how this will simplify workflows for developers working with AI and custom agents.

Congratulations for the launch! I love the seamless integration to SWE-Bench. This is going to really turbocharge coding agent progress.

Is anyone actually using PR Review agents in production, I have found none of the existing ones to be useful?

Wow, SWE-Kit is an awesome tool for developers! Building custom coding agents with AI-native tools is such a cool idea. I can't wait to try it out. What's the most exciting feature for you?

Congratulations on the launch! I have been looking for something like this for a while. I am gonna give it a spin soon. Thank you for creating it!

Congrats @karan_vaidya2 and @soham_ganatra3 on the launch 🚀 Amazing to see that it scores better than Devin on swebench!

@karan_vaidya2 & @soham_ganatra3 congrats on your launch! I feel that it finally brings all puzzle pieces together compared to just using LLMs to improve code / write it faster

Those benchmark numbers against Devin are pretty eye-opening! Been looking for a way to streamline our PR reviews and code maintenance without getting locked into complex setups, and SWE-Kit looks like it could be the answer. Really dig the framework-agnostic approach and how it plays nice with existing tools like GitHub and Jira. The fact that it's open-source is a huge plus - means we can actually tweak it for our specific needs instead of being stuck with another black-box solution. Definitely giving that PR Agent a spin this week! 🛠️