Open SWE: The Open-Source, Cloud-Based Coding Agent That Ships Real Software

Software engineering is evolving faster than any IDE plugin can keep up with. Traditional copilots autocomplete; Open SWE autonomously plans, codes, tests, reviews and raises PRs, all in the cloud. No tether to a laptop. No manual shell babysitting. No context gaps across a large repo.

From messy debug sessions to AI‑powered clarity, Open SWE acts like a real teammate: it reads the codebase, forms a plan, executes in a secure sandbox and reports progress in GitHub issues. Connect a repo in minutes, add a label and watch it work so engineering moves from backlog to merged, fast.

Operation model

• Traditional Coding Assistants: Synchronous in-IDE prompts
• Open SWE: Long‑running, asynchronous, cloud‑hosted agent

Context handling

• Traditional Coding Assistants: Limited file/tab context
• Open SWE: Whole‑repo research and planning with a dedicated Planner

Execution

• Traditional Coding Assistants: Suggests snippets
• Open SWE: Writes code, runs tests, executes shell commands in the Daytona sandbox

Workflow integration

• Traditional Coding Assistants: Minimal GitHub integration
• Open SWE: Native GitHub integration: issue tracking, status, auto PRs

Autonomy

• Traditional Coding Assistants: Human‑driven per prompt
• Open SWE: Multi‑agent flow: Manager - Planner - Programmer - Reviewer

Safety/Quality

• Traditional Coding Assistants: Ship first, fix later
• Open SWE: Plans before coding; Reviewer validates before PR to reduce broken builds

Parallelism

• Traditional Coding Assistants: Single developer’s machine
• Open SWE: Cloud concurrency: multiple tasks in parallel without local resources

Triggering

• Traditional Coding Assistants: Manual prompt in editor
• Open SWE: Web UI or GitHub labels like open-swe or open-swe-auto

Real-world impacts teams care about:

• Fewer review cycles thanks to plan‑then‑review architecture, Engineer‑level PRs with pre‑flight checks.
• Cloud autonomy means no dev workstation drain and no stalled builds.
• Issue‑to‑PR continuity keeps managers and maintainers in the loop automatically.

• Actively developed open-source project: with public repo activity and issues/pull requests flowing under langchain‑ai/open‑swe.
• “Top contributor” effect: Open SWE has been used internally by LangChain to accelerate development on projects like LangGraph and its own repo, where it’s already a top contributor.
• Hosted demo: teams can connect GitHub and kick off tasks in minutes at swe.langchain.com (bring your Anthropic API key).
• Press and community coverage: InfoQ highlights the agent’s multi‑agent architecture, sandbox execution and human‑in‑the‑loop controls.
• License: Open-source; the repository is public under langchain‑ai/open‑swe, designed for forking and extension.

Open SWE is built on LangGraph and deployed on LangGraph Platform for long‑running, stateful agents, with LangSmith used for observability and evaluation of context and prompt design. Its multi‑agent architecture is clear and controllable:

• Manager: orchestrates entry, state and routing.
• Planner: researches repo, proposes a detailed execution plan, supports human review before execution.
• Programmer: executes steps, writes code, runs tests and tools, searches docs.
• Reviewer: validates outputs, loops back for fixes and only then opens a PR.

Runs securely in an isolated Daytona sandbox so it can freely execute shell commands without endangering host environments. Triggers via:

• Web UI with manual/auto modes, including live token and thread insights.
• GitHub label workflows open-swe or open-swe-auto on issues kick off tasks and link results back to issues.

Spec Table

Core framework

• LangGraph (agent orchestration); LangGraph Platform for deployment/persistence

Observability

• LangSmith for tracing/evaluation

Execution

• Daytona cloud sandboxes per session

Interaction modes

• Web UI (swe.langchain.com) and GitHub labels/webhooks

Agents

• Manager, Planner, Programmer, Reviewer

LLM providers

• Anthropic (recommended), with options for OpenAI/Google in local dev envs

Keys required (hosted)

• Anthropic API key for demo; connect GitHub account

Keys required (self‑host)

• LangSmith, Anthropic/OpenAI/Google (at least one), Daytona, GitHub App credentials

CI/CD

• Configurable; docs provide CI configuration guidance

Sample I/O Table

Input (Task/Issue): “Add CI check to ensure commit messages follow Conventional Commits across monorepo.”

• Agent Behavior: Planner drafts steps to modify CI config, add a linter/hook, update docs and tests; Manager handles state; Programmer edits workflows and scripts; Reviewer validates
• Output: Opened PR with updated .github/workflows, scripts and docs; linked to tracking issue with status and plan history

Input (Task/Issue): “Refactor legacy utils to TypeScript with strict types and update consuming modules.”

• Agent Behavior: Planner enumerates files, dependency impacts and migration plan; Programmer executes conversions with tsconfig changes and tests; Reviewer fixes type gaps pre‑PR
• Output: PR with refactors, type fixes and passing tests; issue shows execution logs and plan approvals

GitHub: Offfical public repo with issues, PRs and active commits.

Docs & usage guides: Overview, usage via UI/GitHub, best practices, examples and setup.

Media/briefing: InfoQ coverage summarising design choices and developer controls.

Community dynamics: Open, extensible and built for forks, customise prompts, add internal tools, or modify agent logic to fit enterprise workflows.

Hosted (fastest path):

• Visit swe.langchain.com and sign in.
• Connect GitHub and select target repositories.
• Add the Anthropic API key in settings.
• Start a task from the web UI, or label a GitHub issue open-swe or open-swe-auto.

Local development / self‑hosting:

• Clone the repo: git clone https://github.com/langchain-ai/open-swe.git && cd open-swe.
• Install deps: yarn install.
• Copy env files: cp apps/web/.env.example apps/web/.env and cp apps/open-swe/.env.example apps/open-swe/.env.
• Configure keys:
• Web app: NEXT_PUBLIC_API_URL, LANGGRAPH_API_URL, encryption key, GitHub OAuth, GitHub App info.
• Agent: LangSmith keys, at least one LLM key (Anthropic recommended), Daytona API key, optional tools, GitHub App details, webhook secret, PORT, app URL, encryption key.
• Create a GitHub App and set IDs/secrets/private key.
• Run services following the dev setup guide; integrate CI as desired.

Version: Initial open-source release,

• Date: Aug 2025
• Feature: Public repo, multi‑agent architecture (Manager/Planner/Programmer/Reviewer), GitHub integration, Daytona sandbox, hosted demo at swe.langchain.com

Upcoming (from announcement/considerations):

• Local CLI mode optimised for small, one‑liner fixes, where planning/review are optional complements the cloud async agent for a full spectrum toolchain.
• Continued enhancements to UI observability (token usage, thread inspection), planning UX and scale via LangGraph Platform.

Q1. What is Open SWE, in one line?

An open‑source, cloud‑based, asynchronous coding agent that plans, codes, tests, reviews and opens PRs autonomously, integrated with GitHub.

Q2. How does it integrate with GitHub workflows?

It creates/updates tracking issues with plans and status, triggers from labels like open-swe/open-swe-auto and opens PRs on completion.

Q3. How does it keep code safe and stable?

Every run uses an isolated Daytona sandbox for execution and a Reviewer validates and fixes issues before any PR is opened.

Q4. What powers the agent under the hood?

LangGraph orchestrates multi‑agent flows; LangGraph Platform enables long‑running, persistent deployments; LangSmith provides tracing/evals.

Q5. Can teams self‑host?

Yes, clone the repo, configure environments (LLM keys, Daytona, GitHub App) and follow the development setup docs; enterprise deployment is supported via LangGraph Platform or a custom API server for LangGraph.

Q6. Do I need an LLM key?

Yes, the hosted demo requires an Anthropic API key; local development supports Anthropic and optionally OpenAI/Google.

Try Open SWE Today. Official Docs 

Connect GitHub, add your Anthropic key and ship smarter PRs from issue to merge without babysitting builds. Code with confidence, debug with drive and level up from junior to senior dev by harnessing an agent that plans, tests and reviews before it ships.

Need help with AI transformation? Partner with OneClick to unlock your AI potential. Get in touch today!