Open-SWE

Open-source asynchronous AI coding agent built on LangGraph by LangChain. Positioned explicitly as a free alternative to Google’s Jules (which moved many of its features behind a paywall). Plans, writes, tests, reviews, and opens GitHub pull requests autonomously — all the same surface area as Jules, but locally hostable, free if you bring your own API key (or use OpenRouter’s free tier or Ollama).

• GitHub: langchain-ai/open-swe
• Cloud demo: swe.langchain.com
• Docs: docs.langchain.com/labs/swe/setup/development
• Blog post: blog.langchain.com/introducing-open-swe-an-open-source-asynchronous-coding-agent/
• License: Open source

What it does

Open-SWE is an asynchronous coding agent — it works in the background while you do other things, rather than blocking your IDE on each request. The async framing matters because the typical workflow is:

1. Open a GitHub issue (or use the Open-SWE UI) describing what you want
2. Open-SWE clones the repo, reads the codebase, and produces a step-by-step plan
3. You review the plan in the web UI (or accept it as-is)
4. Open-SWE executes the plan: writes code, runs tests, iterates on failures
5. Open-SWE opens a pull request when done
6. You review and merge

You’re not watching it work. You queue it up and check back later. This is the “Jules-like” interaction model: AI as a remote teammate, not as an inline autocomplete.

Capabilities

• Codebase understanding — clones and reads the repo before planning
• Plan-first execution — produces a step-by-step plan for human review before any code is written
• Multi-file edits — across the entire repo
• Test execution — runs tests as part of the workflow, iterates on failures
• Self-review — has a separate “review” agent that checks the work
• Pull request automation — opens PRs against the linked GitHub repo when done
• Token usage visibility — hover the model dropdown to see input/output token counts and total cost per generation

How to run it

Per the WorldofAI walkthrough:

Cloud option: just use swe.langchain.com (free tier with limits, then paid).

Self-hosted option (the wiki’s recommendation):

git clone https://github.com/langchain-ai/open-swe
cd open-swe
yarn install
# configure two .env files: web app + agent
yarn dev   # starts the LangGraph agent
# in another terminal:
yarn dev   # starts the web app

Then create a GitHub App for repository integration, configure webhooks via ngrok (for local) or your domain (for cloud), and point Open-SWE at your repo.

Backend options:

• Bring your own API key (Anthropic, OpenAI, Google, etc.)
• Use OpenRouter’s free tier for cost-free remote inference
• Use Ollama locally for fully-local operation, no API costs

The local-via-Ollama path is the wiki’s preferred configuration — it’s the most aligned with the open-source model integration thread.

How it compares

	Open-SWE	Google Jules	Claude Code	Cursor	Archon OS
Form factor	Web UI + GitHub integration (cloud or self-hosted)	Cloud web UI	CLI agent harness	Standalone editor	Hosted dashboard
Async vs sync	Async (queue and walk away)	Async	Sync (interactive)	Sync (interactive)	Async (real-time editable task list)
Open source	Yes	No	No (open SWE-bench pipeline only)	No	Yes
Self-hostable	Yes	No	Local CLI	Local editor	Yes
Local model support	Yes (Ollama)	No	Via substitution	Limited	Yes
Underlying framework	LangGraph	Custom Google	Custom Anthropic	Custom	Custom
Underlying model	Bring your own	Gemini 2.5 Pro	Claude (or substitute)	Multi	Bring your own

Open-SWE’s distinctive features:

• Async + open-source + self-hostable + bring-your-own-model — no single competitor matches all four
• Built on LangGraph, which means the agent’s structure is inspectable and customizable in ways closed-source agents aren’t

Caveats from the source

• Worldofai’s framing is enthusiastic (“INSANELY GOOD!” in the title) — discount per the channel’s typical bias
• The cloud demo is the easy path; self-hosting requires GitHub App creation, ngrok setup for webhooks, and dual-yarn-dev terminal management. Not a 5-minute install.
• Jules comparison is the framing — the video keeps coming back to “this is what Jules does, but free.” Useful for setting expectations, but if you don’t already know what Jules does the comparison doesn’t tell you what Open-SWE does on its own merits.
• Demo project in the video is a calendar app — small enough that the autonomous PR-opening worked. Larger codebases will stress-test the planning step.

Where this fits in the wiki

• Agentic coding platform proliferation — Open-SWE is another entry in the fragmenting agentic-coding market. Distinctive for being async + open-source, which is rare in this space.
• Open-source agent frameworks — alongside Hermes Agent, Paperclip, ThePopeBot in the open-source-agent-framework cluster
• LangChain’s product strategy — Open-SWE is the wiki’s first concrete LangChain-shipped product, which justifies the langchain org page

See Also

• langchain — the org behind Open-SWE
• ollama, openrouter — backend options
• claude-code, cursor, augment-agent, firebase-studio, archon-os, deepcode — competing agentic-coding platforms
• saas-death-spiral — the broader market thesis
• hermes-agent, paperclip, thepopebot — adjacent open-source agent frameworks
• Source: Open-SWE walkthrough