Agentless approach to automatically solve software development problems
Top 24.0% on sourcepulse
Agentless provides an agentless approach to automatically solve software development problems, targeting researchers and developers working with automated program repair. It aims to improve efficiency and reduce costs in bug fixing by leveraging large language models without requiring complex agent architectures.
How It Works
Agentless employs a three-phase process: localization, repair, and patch validation. It first localizes faults hierarchically to specific files, classes, functions, and edit locations. Then, it samples multiple candidate patches in diff format for each bug. Finally, it selects regression tests, generates reproduction tests, and re-ranks patches based on test results to submit a final fix. This approach simplifies LLM integration for software engineering tasks.
Quick Start & Requirements
conda create -n agentless python=3.11
, conda activate agentless
), install requirements (pip install -r requirements.txt
), and set PYTHONPATH
.OPENAI_API_KEY
. Full SWE-bench lite reproduction details are in a separate README.Highlighted Details
Maintenance & Community
Licensing & Compatibility
Limitations & Caveats
7 months ago
Inactive