🌾 🥳 🌋 🏰 🌅 🌕 OpenHands Advanced 🌖 🌔 🌈 🏆 👑
MopenHands + Claude-3.7-Sonnet achieves state-of-the-art performance on the SWE-bench leaderboard.
-
Copy the example environment file:
cp .env.example .env
-
Configure your environment variables in
.env:
| Variable | Description | Example |
|---|---|---|
LLM_API_KEY |
Your LLM provider API key | sk-ant-api03-... |
| Variable | Description | Example |
|---|---|---|
SEARCH_API_KEY |
API key for search functionality (Tavily) | tvly-*** |
| Variable | Description | Default | Options |
|---|---|---|---|
LLM_PROVIDER |
LLM provider to use | anthropic |
anthropic, openai, groq |
LLM_MODEL |
Model to call | anthropic/claude-sonnet-4-20250514 |
Provider-specific model names |
LLM_NUM_RETRIES |
How many times to retry a failed completion | 3 |
1-10 |
LLM_CACHING_PROMPT |
Enable prompt/result cache | true |
true, false |
LLM_REASONING_EFFORT |
Agent reasoning effort level | high |
low, medium, high |
| Variable | Description | Default | Options |
|---|---|---|---|
AGENT_MEMORY_ENABLED |
Persist and recall agent memory across runs | true |
true, false |
AGENT_ENABLE_THINK |
Emit the agent's internal THINK steps to the logs | true |
true, false |
AGENT_ENABLE_MCP |
Enable multi-component-planning mode | false |
true, false |
| Variable | Description | Default | Options |
|---|---|---|---|
SANDBOX_PLATFORM |
Target platform architecture | linux/amd64 |
linux/amd64, linux/arm64 |
SANDBOX_ENABLE_GPU |
Enable GPU support in sandbox | false |
true, false |
SANDBOX_RUNTIME_CONTAINER_IMAGE |
Container image for the runtime sandbox | all-hands-ai/runtime:latest |
Any valid container image |
CONTAINER_NAME |
Name given to the Docker container | oha-cli-container |
Any valid container name |
| Variable | Description | Default | Options |
|---|---|---|---|
LOG_LEVEL |
Logging level for the application | INFO |
DEBUG, INFO, WARNING, ERROR, CRITICAL |
LOG_ALL_EVENTS |
Log all events or only top-level events | true |
true, false |
# Optional - Search functionality
SEARCH_API_KEY=tvly-***
# Required - LLM API Key
LLM_API_KEY=
# LLM Configuration
LLM_PROVIDER="anthropic"
LLM_MODEL="anthropic/claude-sonnet-4-20250514"
LLM_NUM_RETRIES=3
LLM_CACHING_PROMPT=true
LLM_REASONING_EFFORT="high"
# System Configuration
SANDBOX_PLATFORM="linux/amd64"
SANDBOX_ENABLE_GPU=false
SANDBOX_RUNTIME_CONTAINER_IMAGE="all-hands-ai/runtime:latest"
# Agent Configuration
AGENT_MEMORY_ENABLED=true
AGENT_ENABLE_THINK=true
AGENT_ENABLE_MCP=false
# Container Configuration
CONTAINER_NAME="oha-cli-container"
# Logging Configuration
LOG_LEVEL="INFO"
LOG_ALL_EVENTS=true📖 For detailed LLM configuration options, visit: LLM Configuration Guide
chmod +x run-open-hands.sh./run-open-hands.shThe script will:
- Load environment variables from
.env - Set up the workspace directory
- Run the Open Hands container with the necessary configurations
- Mount required volumes for workspace and Docker socket
- Connect to the specified LLM provider
- Docker installed and running
- Valid LLM Provider API key
- Proper permissions to access Docker socket
Check it out here to learn how to prompt efficiently.
Create repository-specific Microagents to store the repository knowledge base. Check it out here
Put all development documents to folder reference_docs at root repository folder (Markdown format is recommended). Create index.md to brief the overview of all document files. You can create each index.md file for each sub-folder.