deepseek-ai/deepseek-harness
DeepSeek Harness (dsh) is an open-source agent harness from DeepSeek AI built on an everything-is-a-plugin architecture over the Cordis framework, started from npm with a local web UI.
Awesome AI Agents › AI Developer Tools
SWE-bench/SWE-smith
SWE-smith is a comprehensive toolkit designed to facilitate the training and scaling of software engineering (SWE) agents. It enables users to create an unlimited number of task instances in the style of SWE-bench for any Python repository, allowing for extensive and diverse training data generation. The toolkit supports the generation of trajectories where SWE-agents solve these task instances, providing valuable data for training and improving the agents' software engineering capabilities. Additionally, SWE-smith allows for the training of local language models (LMs) on these trajectories, exemplified by the SWE-agent-LM-32B model, which demonstrates significant performance improvements in software engineering tasks. The project is well-documented, offering detailed guides on installation, creating task instances, and training SWE-agents. It requires Docker for creating execution environments and is primarily developed and tested on Ubuntu 22.04.4 LTS, with no current plans to support Windows or MacOS. SWE-smith is part of a broader ecosystem that includes related resources such as a large dataset of Python task instances, pre-trained models, and training trajectories, all available on the SWE-bench HuggingFace repository. The toolkit is open to contributions and future enhancements, including support for non-Python languages, new bug generation techniques, and advanced training methods. It is licensed under the MIT License and is backed by academic research, as indicated by its arXiv publication. SWE-smith is closely related to other projects like SWE-bench, SWE-agent, SWE-ReX, and sb-cli, forming a robust infrastructure for advancing software engineering automation through AI agents. Overall, SWE-smith provides a scalable and extensible platform for developing and refining software engineering agents, making it a valuable resource for researchers and developers in the field of AI-driven software engineering.
https://github.com/SWE-bench/SWE-smith
Also in AI Developer Tools
Hmbown/Codewhale
Open-source terminal coding agent written in Rust that edits files, runs commands and works toward a goal across many model providers.
TabbyML/tabby
Tabby is a self-hosted, open-source AI coding assistant providing a private and customizable alternative to GitHub Copilot with rich integration and deployment options.
cursor/cursor
Cursor is an AI-powered code editor designed to enhance programming productivity by integrating intelligent code suggestions and AI-driven features directly into the coding workflow.
oraios/serena
Serena is a free and open-source coding agent toolkit that enhances large language models with semantic code retrieval and editing capabilities, enabling autonomous coding workflows directly on your codebase.
QwenLM/qwen-code
Open-source terminal coding agent from the Qwen team with subagents, skills, hooks and MCP support, speaking the OpenAI, Anthropic, Gemini and Qwen API formats plus local model runtimes.
plandex-ai/plandex
Plandex is an open-source AI coding agent designed to plan, execute, and manage large-scale coding tasks and real-world projects with high efficiency and flexibility.
smol-ai/developer
Smol Developer is a library and toolset that embeds a junior developer AI agent into applications to scaffold and generate codebases interactively through human-in-the-loop prompt engineering.