What's New

LangChain released Open SWE, an open-source framework designed to help organizations build internal coding agents that integrate directly into existing developer workflows. The framework emerged from analysis of production systems at Stripe (Minions), Ramp (Inspect), and Coinbase (Cloudbot), which independently converged on similar architectural patterns.

Core Architectural Components

Open SWE provides standardized implementations of five key patterns observed across production deployments:

Agent Harness: Built on LangChain's Deep Agents framework, allowing composition and customization without forking core agent logic. The system supports upgrade paths as improvements flow in from upstream.

Isolated Sandbox Execution: Each task runs in dedicated cloud sandboxes with full shell access and permissions contained within strict boundaries. The framework supports multiple sandbox providers out of the box: Modal, Daytona, Runloop, and LangSmith Sandboxes.

Curated Toolset: Rather than accumulating hundreds of tools, Open SWE ships with a focused set: execute, fetch_url, http_request, commit_and_open_pr, linear_comment, and slack_thread_reply, plus Deep Agents' built-in file and search tools.

Context Engineering: Tasks load context from two sources—an AGENTS.md file at the repository root for organization-wide conventions, and full issue/thread history for task-specific context—reducing the need for exploratory tool calls.

Orchestration: Subagent spawning via the task tool allows complex work to be decomposed into specialized child agents, each with isolated context and focused responsibilities. Middleware hooks enable deterministic control over agent behavior.

Getting Started

Organizations exploring internal coding agents can use Open SWE as a starting point, customizing sandbox backends, tool inventories, and middleware stacks to their specific requirements. The framework integrates with Slack, Linear, and GitHub, meeting developers in existing workflows rather than requiring adoption of new interfaces.

The repository is available on GitHub, built on proven patterns from multiple production deployments across different organizations and industries.