# Contents

LangChain just launched Open SWE, an open-source, async coding agent that reads GitHub issues, plans its tasks, writes and tests code, reviews itself, and opens pull requests. It runs entirely in the cloud, like a real software engineer.

But what sets it apart isn’t just the UX or the open-source license.

It’s the architecture.


Built for the Future of Agents

Open SWE isn’t just another Copilot-style tool. It is a long-running, multi-agent system deployed on the LangGraph Platform with agents that:

  • Read and plan across your entire codebase

  • Execute multi-step workflows over extended time horizons

  • Interrupt themselves for human feedback or direction

  • Review their output and reiterate if needed

Each task can run for an hour or more. That requires real infrastructure.

This is exactly where Daytona comes in.


Why Daytona

Most cloud runtimes are built for humans. Short-lived functions. Stateless interactions. Limited I/O. That does not work when your agent is researching a repo, writing multiple files, running tests, and reasoning about failures for 30 minutes or more.

Daytona is built from the ground up for:

  • Long-running, stateful environments. Each task gets a full system with a persistent state

  • True isolation. Agents can run shell commands, install packages, and execute test suites without human gating

  • Secure, on-demand sandboxes. Purpose-built for autonomous execution with no local dependency

Open SWE uses one Daytona sandbox per task. It is secure, fast to spin up, and completely disposable once the job is done. Like giving each task its own personal VM with full control.


Want to Run Open SWE Yourself

You can. Open SWE is fully open source and extensible. All you need is:

The hosted version runs at swe.langchain.com, but the real power is being able to fork and customize your agent stack.


The Bottom Line

LangChain didn’t just build a coding agent. They built a real software automation system. Long-running. Multi-agent. Cloud-native.

And they didn’t pick Daytona by accident.

Daytona is the only platform purpose-built for agents like Open SWE. Ones that need memory, persistence, isolation, and runtime control. Not autocomplete. Actual engineering.


→ Try Open SWE
→ Fork the repo
→ Start with Daytona

Tags::
  • langchain
  • open swe
  • daytona
  • ai agents
  • cloud runtime
  • coding agents
  • autonomous agents
  • multi-agent systems
  • sandboxing
  • infrastructure
  • langgraph
  • devtools
  • async workflows
  • persistent environments
  • secure execution