Open SWE - An Open-Source Asynchronous Coding Agent
Open SWE is an open-source cloud-based asynchronous coding agent built with [LangGraph](https://docs.langchain.com/oss/javascript/langgraph/overview). It autonomously understands codebases, plans solutions, and executes code changes across entire repositoriesβfrom initial planning to opening pull requests.
>
> **Note: you're required to set your own LLM API keys to use the demo.**
> [!NOTE]
> π¬ Read the **announcement blog post [here](https://blog.langchain.com/introducing-open-swe-an-open-source-asynchronous-coding-agent/)**
# Features
- π **Trigger from Linear, Slack, or GitHub** β mention `@openswe` in a Linear comment, Slack thread, or GitHub PR comment to kick off a task
- π **Instant acknowledgement** β reacts with π the moment it picks up your message so you know it's on it
- π¬ **Message it while it's running** β send follow-up messages mid-task and it'll pick them up before its next step
- π **Run multiple tasks in parallel** β each task runs in its own isolated cloud sandbox, no queuing
- π **GitHub OAuth built-in** β authenticates with your GitHub account automatically, no token setup needed
- π **Opens PRs automatically** β commits changes and opens a draft PR when done, linked back to your Linear ticket
## Installation
### Prerequisites
- Python 3.11+
- [uv](https://docs.astral.sh/uv/) package manager
- [LangGraph CLI](https://langchain-ai.github.io/langgraph/cloud/reference/cli/)
- [ngrok](https://ngrok.com/) (for exposing local webhooks)
### 1. Clone the repo
```bash
git clone https://github.com/langchain-ai/open-swe.git
cd open-swe/apps/agent
```
### 2. Install dependencies
```bash
uv sync
```
### 3. Set up the Linear webhook
In a terminal, start ngrok to get your public URL:
```bash
ngrok http 8000
# e.g. https://xxxx.ngrok.io
```
Then in Linear:
1. Go to **Settings** β **API** β **Webhooks** β **New webhook**
2. Fill in:
- **Label**: `open-swe`
- **URL**: `https://xxxx.ngrok.io/webhooks/linear`
- **Secret**: generate a random string β copy it, you'll need it for `LINEAR_WEBHOOK_SECRET`
3. Under **Data change events**, enable **Comments** β `Create` only
4. Click **Create webhook**
To get your `LINEAR_API_KEY`:
1. Go to **Settings** β **API** β **Personal API keys** β **New API key**
2. Name it `open-swe` and copy the key
### 4. Set environment variables
Create a `.env` file in `apps/agent/` with the following:
```bash
# LangSmith
LANGSMITH_API_KEY_PROD="" # Your LangSmith API key
LANGCHAIN_TRACING_V2="true"
LANGCHAIN_PROJECT=""
# LLM
ANTHROPIC_API_KEY="" # Anthropic API key (recommended default provider)
# GitHub OAuth (via LangSmith agent auth)
GITHUB_OAUTH_PROVIDER_ID="" # GitHub OAuth provider ID from LangSmith
X_SERVICE_AUTH_JWT_SECRET="" # Secret for service JWT tokens
# GitHub App (Bot)
GITHUB_APP_ID="" # GitHub App ID
GITHUB_APP_PRIVATE_KEY="-----BEGIN RSA PRIVATE KEY-----
...
-----END RSA PRIVATE KEY-----
"
GITHUB_APP_INSTALLATION_ID="" # GitHub App installation ID
# GitHub Webhook
GITHUB_WEBHOOK_SECRET="" # Secret for verifying GitHub webhooks
# Linear
LINEAR_API_KEY="" # Linear API key (from step 3)
LINEAR_WEBHOOK_SECRET="" # Secret you set when creating the webhook (from step 3)
# Slack (optional)
SLACK_BOT_TOKEN=""
SLACK_BOT_USER_ID=""
SLACK_BOT_USERNAME=""
SLACK_SIGNING_SECRET=""
# Sandbox
DEFAULT_SANDBOX_TEMPLATE_NAME="" # LangSmith sandbox template name (uses default if not set)
# Token encryption
TOKEN_ENCRYPTION_KEY="" # 32-byte url-safe base64 key for encrypting GitHub tokens
```
### 5. Run the agent
In one terminal, start the LangGraph dev server:
```bash
uv run langgraph dev --no-browser
```
In a second terminal, start the webhook server:
```bash
make run
```
The LangGraph server runs on `http://localhost:2024` and the webhook server on `http://localhost:8000`.
### 6. Verify it works
Comment `@openswe` on any Linear issue. You should see:
- A π reaction on your comment within a few seconds
- A new run appear in your LangSmith project
---
## Usage
Open SWE can be used in multiple ways:
- π **From Linear**. Mention `@openswe` in a comment on any Linear issue and describe the task you want it to perform (e.g. `@openswe fix the login bug described above`). The agent will pick up the issue context along with your instructions and start working on it.
- π **GitHub (for Open SWE-generated PRs)**. Once Open SWE completes implementation, it pushes changes to a branch `open-swe/` and opens a **draft pull request** linking back to the originating Linear issue. From there, you can review the code, request changes, and merge when ready. 