Open SWE: An Open-Source Framework for Internal Coding Agents

Open SWE is an open-source framework built on Deep Agents and LangGraph for creating customizable internal coding agents that fit existing engineering workflows.

Visit Website
Open SWE: An Open-Source Framework for Internal Coding Agents

Introduction

Overview

Open SWE is an open-source framework for organizations building internal coding agents. Built on Deep Agents and LangGraph, it packages architectural patterns observed in production systems into a customizable starting point for engineering teams.

The framework is designed around agents that work alongside developers through familiar tools rather than requiring a separate interface. Organizations can adapt its components to their own environments, policies, and workflows.

Key Architecture Patterns

Open SWE reflects several recurring choices found in internal coding-agent deployments:

  • Isolated cloud sandboxes let agents execute commands within strict boundaries, limiting the impact of mistakes.
  • Curated toolsets give agents maintained capabilities selected for useful engineering work.
  • Rich startup context can be drawn from issues, conversations, and pull requests before work begins.
  • Subagent orchestration breaks complex assignments into focused tasks with separate contexts.
  • Workflow integrations make agents accessible through communication and development systems teams already use.

Framework Design

Open SWE composes its agent harness on the Deep Agents framework instead of forking an existing agent or rebuilding the core from scratch. This structure allows improvements in areas such as context management, planning, and token usage to be incorporated without discarding organization-specific work.

Custom tools, prompts, and workflows can remain configuration around the underlying agent logic. LangGraph provides the foundation for controlling complex agent systems, while Open SWE focuses those capabilities on internal software-engineering use cases.

Use Cases

Engineering organizations can use Open SWE as a foundation for agents that receive development context, operate in isolated environments, and coordinate specialized work. Its architecture is suited to teams exploring coding agents that participate in established issue, communication, and code-review workflows while retaining control over their own customizations.