Open Swe
An Open-Source Asynchronous Coding Agent
Last verified:
What is Open Swe?
Open SWE is an open-source cloud-based asynchronous coding agent built with LangGraph that autonomously understands codebases, plans solutions, and executes code changes across entire repositories from initial planning to opening pull requests. It represents a shift away from real-time copilot assistants toward more autonomous, long-running agents that integrate directly with a developer's existing workflows, acting like another member of the engineering team.
The tool operates through three specialized LangGraph agents: a Manager Graph that orchestrates user interactions and coordinates between other graphs, a Planner Graph that analyzes requirements and creates detailed execution plans, and a Programmer Graph that executes code changes based on approved plans. It also includes a Reviewer agent that checks outputs for errors before any pull request is created, aiming to reduce broken builds and repeated review cycles.
Key features include dedicated planning with the ability to accept, edit, or reject proposed plans before execution, human-in-the-loop controls allowing developers to send messages while the agent is running for real-time feedback, parallel execution of unlimited tasks in a cloud sandbox environment, and end-to-end task management that automatically creates GitHub issues and pull requests that close issues when implementation is complete.
Open SWE is designed for software developers, engineering teams, and developers who need to handle complex, high-context, longer-term software development tasks. It connects directly to GitHub repositories and works on tasks that can run for an hour or more, making it ideal for bug fixes, feature implementations, code refactoring, and other multi-step workflows that require deep codebase understanding.
The agent runs entirely in a secure, isolated Daytona sandbox in the cloud, allowing it to execute shell commands freely without risking the host environment. This architecture enables it to operate without consuming local resources and handle multiple tasks in parallel simultaneously.
Open Swe pricing
Pricing model: Freemium
Open SWE is fully open source under the MIT license and free to use. The hosted version runs at swe.langchain.com as a public demo where users are required to set their own LLM API keys. There is a free tier available with pricing plans coming soon. For self-hosting, users need a model API key (Anthropic, OpenAI, etc.), a Daytona API key, and the Open SWE GitHub repo. The tool itself is free but users pay for their own LLM API usage and Daytona sandbox costs.
Open Swe pros
- Open source and MIT licensed for free use and customization
- Cloud-based execution doesn't consume local resources
- Autonomous end-to-end task management from planning to pull requests
- Dedicated planning step for deep codebase understanding
- Human-in-the-loop controls for real-time feedback during execution
- Ability to accept, edit, or reject plans before execution
- Parallel execution of unlimited simultaneous tasks
- Secure isolated Daytona sandbox environment for each task
- Direct GitHub integration via webhooks and issue labels
- Automatic GitHub issue creation and pull request generation
- Double texting capability to send messages while agent is running
- Multi-agent architecture with Manager, Planner, Programmer, and Reviewer
- Reviewer agent checks outputs for errors before creating pull requests
- Self-hosting option available for enterprise use
- Claude Opus 4.1 support via open-swe-max labels for complex tasks
Open Swe cons
- Requires users to set their own LLM API keys
- Requires Daytona API key for setup
- Cloud-based so needs internet connection for all operations
- Tasks can run for an hour or more which may be slow for quick fixes
- Not a real-time IDE copilot for immediate code suggestions
- Self-hosting requires writing an API server for LangGraph
- May be overkill for simple one-line code changes
- Learning curve for setting up development environment
- Depends on Anthropic/OpenAI model API costs
- GitHub webhook setup required for automated workflows
- May require human review before merging generated pull requests
- Not suitable for tasks requiring local environment access
- Cloud sandbox may have limitations on certain system commands
- Limited to languages and frameworks it has been trained on
- Documentation primarily targets developers comfortable with LangGraph
Frequently asked questions about Open Swe
What is Open SWE?
Open SWE is an open-source cloud-based asynchronous coding agent built with LangGraph. It autonomously understands codebases, plans solutions, and executes code changes across entire repositories from initial planning to opening pull requests. Unlike real-time copilot assistants, it operates as a long-running autonomous agent that integrates directly with developer workflows.
How does Open SWE work?
Open SWE operates through three specialized LangGraph agents: the Manager Graph orchestrates user interactions and coordinates between other graphs, the Planner Graph analyzes requirements and creates detailed execution plans, and the Programmer Graph executes code changes based on approved plans. A Reviewer agent also checks outputs for errors before pull requests are created. Each task runs in a secure, isolated Daytona sandbox in the cloud.
Is Open SWE free to use?
Yes, Open SWE is fully open source under the MIT license and free to use. The hosted version is available as a public demo at swe.langchain.com. However, users need to provide their own LLM API keys (from Anthropic, OpenAI, etc.) and a Daytona API key, which may have associated costs.
How do I integrate Open SWE with GitHub?
You can start Open SWE tasks directly from GitHub issues by adding the label 'open-swe' or 'open-swe-auto'. The '-auto' variant causes Open SWE to automatically accept the plan without intervention. For enhanced performance on complex tasks, use 'open-swe-max' or 'open-swe-max-auto' labels which utilize Claude Opus 4.1 for both planning and programming.
Can I send messages to Open SWE while it's running?
Yes, Open SWE has human-in-the-loop controls that allow you to send it messages while it's running during both planning and execution steps. This feature, called 'double texting', allows you to give real-time feedback and instructions without having to interrupt or restart the process.
Can I self-host Open SWE for enterprise use?
Yes, you can self-host Open SWE. You need to either write an API server for LangGraph or use LangChain's self-hosting offering if using the LangGraph Platform to deploy. All you need is a model API key (Anthropic, OpenAI, etc.), a Daytona API key, and the Open SWE GitHub repo.
What types of tasks is Open SWE designed for?
Open SWE is designed for high-context, longer-term tasks such as bug fixes, feature implementations, code refactoring, and multi-step workflows that require deep codebase understanding. Tasks can run for an hour or more, making it ideal for complex software development work rather than quick one-line changes.
How does Open SWE handle task management?
Open SWE provides end-to-end task management by automatically creating GitHub issues for tasks and creating pull requests that close the issue when implementation is complete. You can create, manage, and execute Open SWE tasks from the web application UI or directly from GitHub issues using labels.
What makes Open SWE different from other coding assistants?
Open SWE represents a shift from real-time copilot assistants toward autonomous, long-running agents. Unlike IDE-based assistants focused on short interactions, Open SWE connects directly to GitHub repositories, works like another engineering team member, operates entirely in the cloud without consuming local resources, and can run multiple tasks in parallel.
How do I get started with Open SWE?
You can try the public demo at swe.langchain.com (you'll need to set your own LLM API keys). For local development, see the Open SWE documentation at docs.langchain.com/labs/swe/index which covers setup including monorepo structure, dependencies, and authentication. You can also fork the GitHub repository at github.com/langchain-ai/open-swe to customize your agent stack.