by soham123 on 11/6/24, 2:47 PM with 0 comments
I built SWE-Kit, LLM toolkit (Function callable tools) which makes building agents specialised in coding like Devin very easy.
I noticed a typical pattern while building local agents: creating & perfecting LLM tools to interact with system or codebase was the repeated and time-consuming. We created a layer that simplifies building agents that can interact with code, file system, git, shell and allows you to quickly solve for a wide variety of coding agent use cases.
Aren’t there open coding agents already? Well, yes, but most folks would want to solve their specific use case like a large refactor and current coding agents aren’t customisable to your specific use case or aren’t meant to be molded to different workflows.
The idea is to provide a library of tools so you can build software engineering agents with a few lines of code in agentic framework of your choice.
We have solved following hard parts for everyone - - 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 Local, Docker, Fly.io, E2b, AWS Lambda (soon!)
Is this the 10x Coding Agent I was looking for?
No this is not a coding agent but allows you to build your custom coding agent in framework of your choice.
We have created some templates to get started quickly though: - 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.
To better showcase the SWE kit's capability, we tested it on [swebench.com](https://www.swebench.com/), the benchmark for testing coding agents. It scored 48.60%, whereas Devin scored only 13.86%.
If you end up using this, please do provide feedback and if you need help building coding agent feel free to reach out to us
I (Soham) & my cofounder Karan are both active on this thread to answer any questions!