LangChain Releases Open SWE Internal Coding Agent

LangChain announced the release of Open SWE, an open-source, MIT-licensed framework designed for building customizable internal software engineering agents. Built on Deep Agents and LangGraph, the project captures architectural choices observed in proprietary tools used at major tech firms, such as Stripe's Minions, Ramp's Inspect, and Coinbase's Cloudbot. Open SWE allows developers to execute code safely inside isolated cloud sandboxes, offering default support for providers like Modal, Daytona, Runloop, and LangSmith.

The framework replicates common production patterns highlighted in research by Kishan Dahya. Rather than relying on hundreds of uncurated utilities—such as the roughly 500 tools managed by Stripe's engineering team—Open SWE pairs a curated toolset with rich context gathered from repository AGENTS.md files, Linear issues, Slack threads, and GitHub pull requests. Its environment provides built-in tools like read_file, write_file, edit_file, ls, glob, grep, write_todos, and task, alongside customizable tools like http_request, fetch_url, commit_and_open_pr, linear_comment, and slack_thread_reply.

Under the hood, Open SWE leverages the Deep Agents harness and defaults to anthropic:claude-opus-4-6, though practitioners can swap in any model or sandbox backend. Orchestration combines agentic planning with deterministic middleware hooks, including ToolErrorMiddleware, check_message_queue_before_model, and open_pr_if_needed. These mechanisms ensure incoming messages from Slack or Linear get injected mid-run, while automatically creating a pull request if the agent completes work without opening one.

For engineering practitioners, Open SWE offers a practical blueprint for deploying production-ready assistants directly into everyday developer tools without requiring team members to adopt new interfaces or fork agent code bases from scratch. By standardizing isolated execution boundaries and subagent delegation, organizations can benefit from foundation-level improvements in context management and planning while keeping full control over internal tooling and validation gates.