Open SWE: An Open-Source Framework for Internal Coding Agents
Built on Deep Agents and LangGraph, Open SWE provides the core architectural components for internal coding agents.
This article introduces Open SWE, an open-source framework for building internal coding agents. It outlines the architectural patterns observed in systems developed by companies like Stripe, Ramp, and Coinbase, and provides a customizable foundation for deploying AI agents in engineering workflows.
What happened
The article announces the release of Open SWE, an open-source framework designed to replicate architectural patterns from internal coding agents developed by engineering teams at Stripe, Ramp, and Coinbase. It provides a customizable foundation for building AI agents that integrate with developer workflows and prioritize security and scalability.
Why it matters
For AI developers and engineering teams, Open SWE reduces the complexity of building internal coding agents by offering a standardized, extensible framework. It addresses challenges like secure execution, context management, and workflow integration, enabling organizations to focus on customization rather than foundational development.
- Open SWE provides a customizable foundation for internal coding agents, reducing the need to build from scratch.
- Isolated sandboxes and curated tools enhance security and maintainability of AI-driven workflows.
- Integration with Slack, Linear, and GitHub ensures seamless adoption within existing developer ecosystems.
- The framework's composability allows for upgrades and customizations without forking core logic.
- Organizations can extend the toolset and middleware to align with internal processes and safety requirements.
The bigger picture
Open SWE enables engineering teams to deploy AI agents more efficiently by leveraging proven architectural patterns. Its modular design allows organizations to adapt components to their infrastructure, reducing development time and improving reliability in production environments.
Open SWE offers a standardized, open-source framework for building internal coding agents, drawing from production-tested patterns. It empowers engineering teams to focus on customization while benefiting from secure, scalable, and extensible AI workflows.
Technical details
- Deep Agents and LangGraph as the underlying framework for agent composition and orchestration.
- Isolated cloud sandboxes with full permissions for task execution, preventing production system interference.
- Curated toolset (e.g., execute, fetch_url, commit_and_open_pr) for controlled functionality.
- AGENTS.md file for repository-wide context injection and system prompt customization.
- Subagent spawning via the task tool, with isolated context and middleware hooks for deterministic behavior.
- Middleware mechanisms like ToolErrorMiddleware and open_pr_if_needed for error handling and safety nets.
FAQ
What is Open SWE?
Open SWE is an open-source framework for building internal coding agents, designed to replicate architectural patterns from production systems developed by companies like Stripe, Ramp, and Coinbase.
What are the key features of Open SWE?
Key features include isolated cloud sandboxes, curated toolsets, subagent orchestration, integration with developer tools (Slack, Linear, GitHub), and deterministic middleware for error handling and safety.
How does Open SWE improve security?
It uses isolated execution environments to contain errors, limits tool access to a curated set, and includes validation mechanisms like linters and PR safety nets to ensure reliable outcomes.
Can organizations customize Open SWE?
Yes, Open SWE allows customization of sandbox providers, models, tools, triggers, system prompts, and middleware to fit specific organizational needs.
What is the role of Deep Agents in Open SWE?
Deep Agents provides the foundation for agent composition, context management, and subagent orchestration, enabling scalability and maintainability of the framework.
This summary and analysis were generated with AI from the original article and may contain errors. Read the original for the full story.
Read the full story
Originally published by LangChain Blog on Oct 13, 2026.
Comments
Sign in to join the discussion.