Open SWE is LangChain's open-source framework for building autonomous coding agents — the first MIT-licensed system that captures the architectural patterns Stripe (Minions), Ramp (Inspect), and Coinbase independently converged on for internal coding agents. Built on LangGraph and Deep Agents, it ships with sandbox isolation, Slack/Linear/GitHub invocation, and a multi-agent Manager/Planner/Programmer/Reviewer pipeline.
Architecture Deep Dive → Open SWE Architecture Breakdown — multi-agent architecture, sandbox providers, workflow integration, and comparison with Stripe Minions.
Why It's in Assess
Open SWE launched March 17, 2026 and has grown to 8.7K+ GitHub stars rapidly. The architecture is well-designed — multi-agent orchestration with isolated sandboxes is the pattern that keeps proving itself at enterprise scale. However, the framework is less than two weeks old. No independent production adoption data exists yet outside LangChain's own internal use. The patterns are sound; the implementation needs time to prove itself.
How It Works
Open SWE operates like another engineer on your team: it researches a codebase, creates a detailed execution plan, writes code, runs tests, reviews its own work for errors, and opens a pull request when finished.
The multi-agent pipeline:
- Manager — entry point that handles user interactions and routes tasks
- Planner — analyzes the request, researches the codebase, and proposes a strategy
- Programmer — writes code, runs tests, iterates on failures
- Reviewer — checks outputs for errors before any PR is created
Key Differentiators
- Sandbox isolation — each task runs in its own cloud Linux environment (Modal, Daytona, Runloop, or LangSmith). Multiple tasks can run in parallel.
- Workflow integration — invoke from Slack (mention the bot), Linear (comment @openswe), or GitHub (tag @openswe on PRs)
- Human-in-the-loop — interrupt mid-task, review/edit plans before execution, "double text" new instructions while the agent works
- Curated toolset — ~15 tools (vs. Stripe's ~500), covering shell, web, Git, and integrations
- Model-agnostic — defaults to Claude Opus 4.6 but supports any LLM provider
When to Use Open SWE vs. Copilots
Use autonomous agents like Open SWE when the task spans multiple files, requires research and planning, can run while you work on something else, involves extensive testing, or handles repetitive refactoring. Use copilots (inline completions, chat) for quick edits, single-file changes, and interactive exploration.
Key Characteristics
| Property | Value |
|---|---|
| Provider | LangChain Inc. |
| License | MIT |
| GitHub | langchain-ai/open-swe |
| Stars | 8.7K+ |
| Built on | LangGraph, Deep Agents |
| Default model | Claude Opus 4.6 |
| Sandbox providers | Modal, Daytona, Runloop, LangSmith |
| Launched | March 17, 2026 |