Summary Live-SWE-agent is the first live, runtime self-evolving software engineering agent. It dynamically expands and revises its capabilities while actively addressing real-world software engineering issues. This project provides a unified, open platform for fair, apples-to-apples benchmarking of LLM-based agents, moving beyond proprietary scaffolds.

How It Works The core innovation treats software agents as systems capable of self-modification. Live-SWE-agent leverages LLMs' intrinsic ability to extend or alter their own behavior at runtime. This dynamic adaptation allows the agent to improve its problem-solving strategies and capabilities on-the-fly as it engages with software engineering tasks, offering a novel approach to agent autonomy and evolution.

Quick Start & Requirements Setup requires installing the mini-swe-agent framework first, followed by using a custom Live-SWE-agent configuration file (config/livesweagent.yaml). Specific hardware, OS, or detailed dependency requirements beyond mini-swe-agent are not detailed in the provided README.

Highlighted Details

• Achieves state-of-the-art performance on SWE-bench benchmarks, including a 79.2% solve rate on SWE-bench Verified (using Claude Opus 4.5) and 45.8% on SWE-Bench Pro.
• Demonstrates superior performance compared to proprietary agent scaffolds and enables fair, open comparisons for LLM releases.
• Version 1.0.0 provides access to complete trajectories, patches, and results via Hugging Face datasets.

Maintenance & Community The provided README does not detail specific community channels (e.g., Discord, Slack), notable contributors, sponsorships, or roadmap information.

Licensing & Compatibility License information is not specified in the provided README text. Consequently, compatibility for commercial use or closed-source linking cannot be determined from this information.

Limitations & Caveats No explicit limitations, alpha status, or known bugs are mentioned in the provided README. The project's capabilities are presented as fully realized in version 1.0.0.