Open SWE is an open-source asynchronous coding agent designed for autonomous code modification and repository management. It assists developers by understanding codebases, planning solutions, and executing code changes, ultimately automating the creation of pull requests.

How It Works

Open SWE leverages LangGraph to orchestrate a multi-step process involving planning, execution, and integration with GitHub. Its architecture emphasizes a dedicated planning phase for deep codebase understanding and allows for human-in-the-loop feedback during execution. The system supports parallel task execution in a cloud-based sandbox environment and automates the creation of GitHub issues and pull requests.

Quick Start & Requirements

• UI Demo: Available at https://demo.openswe.ai/ (requires user-provided LLM API keys).
• Local Setup: Refer to the documentation for local setup instructions.
• Prerequisites: Requires LLM API keys (e.g., OpenAI, Anthropic). Specific model requirements (e.g., Claude Opus 4.1 for open-swe-max labels) are noted for enhanced performance.

Highlighted Details

• Autonomous planning with user acceptance/editing of proposed plans.
• Real-time human-in-the-loop feedback during task execution.
• Parallel execution capabilities in a cloud sandbox.
• End-to-end task management, including GitHub issue creation and pull request generation.

Maintenance & Community

• Project is actively developed by langchain-ai.
• Announcement blog post and video are available.
• Documentation is provided.

Licensing & Compatibility

• The repository is licensed under the Apache-2.0 license.
• This license is permissive and generally compatible with commercial and closed-source applications.

Limitations & Caveats

The public demo requires users to supply their own LLM API keys. Enhanced performance for complex tasks is tied to specific, potentially costly, LLM models.