Skip to content
Get packs

Guide · Open source

How to use open-swe: setup, workflow and limits

Updated · written with AI from the project's own documentation and checked against our catalogue data

Open SWE is an open-source coding agent from LangChain that investigates a codebase, makes changes in an isolated sandbox, validates them and opens a pull request. This guide explains what it does, how to get the code, how it is configured and where it is not a fit. By the end you will know whether it suits a team that wants to self-host an agent.

Open SWElangchain-ai/open-swe · ★ 11.2k · MIT

Quick start

  1. 1Clone the repository. Run git clone https://github.com/langchain-ai/open-swe to get the source code.
  2. 2Read the installation guide. Open docs/INSTALLATION.md in the cloned repository and follow its deployment steps.
  3. 3Check the runtime requirement. Confirm you have Managed Deep Agents, or plan to fork and adapt the code to your own runtime.
  4. 4Review the customization guide. Read docs/CUSTOMIZATION.md to pick models, sandbox provider, tools and skills.
  5. 5Set GitHub access and approvals. Limit repositories, credentials and integrations, and set approval policies for your deployment.
  6. 6Start a task. Trigger work from the dashboard, a GitHub issue or PR conversation, or Slack, then review the resulting pull request.

What Open SWE does

Open SWE turns engineering work into a repeatable process: it investigates a repository, implements changes, validates them and delivers a pull request. It also reviews pull requests, monitors CI and responds to feedback.

  • Build: investigate repositories, edit code, run focused validation and open or update pull requests.
  • Review: run on-demand or opt-in automatic reviews and publish findings to GitHub.
  • Investigate: use read-only PR chat to understand a change without implementing anything.
  • Operate: schedule recurring tasks and monitor opted-in PRs with /baby-sit.
  • Customize: choose models, reasoning effort, instructions, skills, integrations and sandbox providers.

Get the source code and install Open SWE

The README's install method is to clone the repository. It lists no separate prerequisites, but the project's installation guide (docs/INSTALLATION.md) covers deployment. Clone the repository with the command below, then follow that guide.

  • Read the README and docs/INSTALLATION.md before deploying.
  • Deploying as-is requires Managed Deep Agents; otherwise plan to fork and adapt the code to your own runtime.
git clone https://github.com/langchain-ai/open-swe

Understand how the agent is built

Open SWE is built on Deep Agents, which supplies planning, filesystem, shell, skills and subagent primitives. LangGraph provides durable execution and thread state. Open SWE adds engineering tools, integrations, authorization and user interfaces.

The graph entrypoints are declared in langgraph.json. Cloud coding runs in persistent, per-thread Linux sandboxes, and an unreachable sandbox is not silently replaced.

Start work from the dashboard, GitHub or Slack

You can start and continue work from the dashboard, from GitHub issues and PR conversations, or from Slack. Linear is supported for issue-comment triggers and replies through a configured Linear MCP connection. Follow-ups in a cloud coding thread reuse that thread's context and sandbox, and independent threads can run in parallel.

Configure GitHub access, integrations and approvals

Access is the part to review most carefully. Coding sandboxes normally receive installation-wide GitHub App access, while some workflows use narrower repository scopes. Workspace repository bindings control routing and preloaded checkouts, but are not a separate credential boundary.

Approval prompts guard detected Git pushes, not every possible shell or API write. Reviewers are instructed not to commit or push, and PR chat excludes mutation tools.

  • Use least-privilege credentials and restrict repositories and integrations.
  • Configure MCP tool scope and credentials via the customization guide.
  • Tailor approval policies to your deployment.
  • Sandboxes have powerful tools and may have network access.

Choose a sandbox and understand local execution

LangSmith is the default sandbox and tracing provider. Other providers and local execution are configurable, as described in docs/CUSTOMIZATION.md. The README is explicit that local execution does not provide cloud sandbox isolation, so use it only where that trade-off is acceptable. PR chat does not need a sandbox.

What people use it for

  • Have an agent investigate a codebase and open a pull request for a routine change.
  • Run automatic or on-demand reviews that post findings to GitHub.
  • Ask read-only questions about a pull request without letting the agent change code.
  • Schedule recurring engineering tasks and monitor opted-in PRs for CI failures.

Frequently asked questions

Is open-swe free to use?

Open SWE is released under the MIT License, which covers its own code. The README does not list prices, and it notes that deploying as-is requires Managed Deep Agents, so check that dependency before assuming the setup is free to run.

What are the requirements for open-swe?

The README lists no prerequisites for the clone-and-build install. Practically you will need access to a Git repository and a GitHub App setup, and the README says deploying as-is requires Managed Deep Agents or a fork adapted to your own runtime.

How do I install open-swe?

Clone the repository with git clone https://github.com/langchain-ai/open-swe, then follow docs/INSTALLATION.md for deployment. The README describes this as a self-deployment for teams that can run their own infrastructure.

Is open-swe ready for production?

The README says it is under active development, with expected breaking changes, and that correctness, stability and compatibility are not guaranteed. Treat it as a reference implementation rather than a stable product.

Can I contribute issues or patches to open-swe?

No, not at this time. The README says the project is not accepting issues or external contributions, though you are welcome to explore and fork the code.

Is open-swe a good alternative to hosted coding agents?

It fits teams that want a self-hosted agent they can inspect and adapt. Teams that want a ready-to-use hosted product, or strong isolation for local execution, should look elsewhere, since the README states neither is provided.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Open SWE suits engineering teams and LangChain or LangGraph developers who want a self-hosted, inspectable coding agent for routine pull request work. Teams that want a ready-made hosted product, stable releases, or the ability to contribute upstream should choose something else.

Install options & downloads