Open SWE
Open SWE is an open-source asynchronous coding agent built by LangChain. It autonomously handles GitHub issues by researching codebases, planning tasks, writing and testing code, self-reviewing, and opening pull requests — acting like a full-time software engineer on your team.
Overview
Open SWE (Open Software Engineering) is an open-source, cloud-native, asynchronous AI coding agent developed by LangChain. It functions as an autonomous teammate that takes GitHub issues (or tasks from a web UI, Slack, or Linear), researches the codebase, creates execution plans, writes code, runs tests, performs self-review, and opens polished pull requests.
Built on LangGraph and Deep Agents, Open SWE captures the internal coding agent architecture used by companies like Stripe, Coinbase, and Ramp, and makes it available to everyone under the MIT license.
Key Features
- Asynchronous & Long-Running: Handles complex, multi-hour or multi-day tasks without blocking.
- Multi-Agent Architecture: Includes Manager, Planner, Programmer, and Reviewer sub-agents for robust execution.
- Cloud Sandbox Execution: Secure, isolated environments for code running and testing (supports Daytona and similar providers).
- Native Integrations: Trigger via GitHub issues/labels, Slack mentions, Linear comments, or custom web UI.
- Automatic PR Creation: Generates well-documented pull requests with changes, tests, and review notes.
- Self-Review & Iteration: The agent reviews its own work and iterates until quality standards are met.
- Fully Customizable: Easy to fork, modify prompts, add internal tools, or adapt to your tech stack.
- Hosted Demo: Try it at swe.langchain.com with your own model key.
How It Works
- Trigger: Create a GitHub issue and add the
open-sweoropen-swe-autolabel, or submit via the web UI. - Planning: The agent explores the repository, understands context, and builds a detailed step-by-step plan.
- Execution: Uses cloud sandboxes to write, test, and debug code.
- Review: Dedicated reviewer agent checks for errors and suggests improvements.
- Output: Opens a PR with full context, or updates the issue with status.
Use Cases
- Resolving complex GitHub issues autonomously.
- Implementing new features or bug fixes in large codebases.
- Internal tool development and maintenance for engineering teams.
- Accelerating open-source contributions or enterprise codebase modernization.
- Building custom internal coding agents tailored to company-specific workflows and security requirements.
Getting Started
- Clone the repository:
git clone https://github.com/langchain-ai/open-swe.git - Follow the detailed Installation Guide for local setup, GitHub App creation, LangSmith tracing, and production deployment.
- Try the hosted version at swe.langchain.com (requires Anthropic or OpenAI API key).
- Documentation is available in the repo under
/apps/docs.
Why Open SWE?
Unlike real-time copilots (Copilot, Cursor, etc.), Open SWE is designed for autonomous, long-running workflows. It shifts the paradigm from “assist me while I code” to “here’s a task — handle it like a senior engineer.”
It is production-ready yet fully extensible, making it ideal for teams wanting to deploy their own secure, internal AI software engineers.
Technical Stack
- Core: LangGraph + Deep Agents
- Language: Python
- Integrations: GitHub, Slack, Linear, cloud sandboxes
- Tracing & Observability: LangSmith
- License: MIT
Open SWE represents the next evolution of AI agents in software engineering — fully open, customizable, and ready for real-world production use.