This guide will help you install and start using crew-opencode for multi-agent orchestration.
- Prerequisites
- Installation
- Configuration
- Your First Workflow
- Understanding SOPs
- Working with Agents
- Advanced Usage
- Troubleshooting
Before you begin, ensure you have:
-
API Keys for your chosen LLM providers:
- Anthropic API - For Claude models (PM, TA, QA)
- OpenAI API - For GPT models (Design)
- Google AI - For Gemini models (FE)
-
Runtime (only for npm installation):
-
Optional: OpenCode CLI for plugin integration
Download the binary for your platform and add it to your PATH:
macOS (Apple Silicon - M1/M2/M3)
curl -L https://github.com/sehyun0518/crew-opencode/releases/download/v1.0.0/crew-opencode-macos-arm64 -o crew-opencode
chmod +x crew-opencode
sudo mv crew-opencode /usr/local/bin/macOS (Intel)
curl -L https://github.com/sehyun0518/crew-opencode/releases/download/v1.0.0/crew-opencode-macos -o crew-opencode
chmod +x crew-opencode
sudo mv crew-opencode /usr/local/bin/Linux (x64)
curl -L https://github.com/sehyun0518/crew-opencode/releases/download/v1.0.0/crew-opencode-linux-x64 -o crew-opencode
chmod +x crew-opencode
sudo mv crew-opencode /usr/local/bin/Windows (x64)
- Download crew-opencode-windows-x64.exe
- Add the directory to your PATH or run directly
Verify Installation
crew-opencode --version
# Should output: 1.0.0# Configure npm to use GitHub Packages
echo "@sehyun0518:registry=https://npm.pkg.github.com" >> ~/.npmrc
# Install globally
npm install -g @sehyun0518/crew-opencode
# Verify installation
crew-opencode --versionbunx @sehyun0518/crew-opencode installSet your API keys as environment variables:
# Add to ~/.bashrc, ~/.zshrc, or ~/.profile
export ANTHROPIC_API_KEY="your-anthropic-key-here"
export OPENAI_API_KEY="your-openai-key-here"
export GOOGLE_API_KEY="your-google-key-here"Don't have all API keys? You can disable agents you don't need (see Configuration File below).
crew-opencode works with default configuration, but you can customize it:
Global Configuration (applies to all projects):
mkdir -p ~/.opencode/crew-opencodeCreate ~/.opencode/crew-opencode/config.json:
{
"version": "1.0.0",
"crew": {
"pm": {
"enabled": true,
"model": "claude-opus-4-5",
"apiKey": "${ANTHROPIC_API_KEY}"
},
"ta": {
"enabled": true,
"model": "claude-sonnet-4.5",
"apiKey": "${ANTHROPIC_API_KEY}"
},
"fe": {
"enabled": true,
"model": "gemini-3-pro",
"apiKey": "${GOOGLE_API_KEY}"
},
"design": {
"enabled": false,
"model": "gpt-5.2-medium",
"apiKey": "${OPENAI_API_KEY}"
},
"qa": {
"enabled": true,
"model": "claude-haiku-4.5",
"apiKey": "${ANTHROPIC_API_KEY}"
}
},
"sop": {
"feature": { "enabled": true },
"bugfix": { "enabled": true },
"refactor": { "enabled": true }
},
"incidentReport": {
"enabled": true,
"outputDir": ".opencode/crew-opencode/incidents"
}
}Local Configuration (project-specific):
mkdir -p .opencode/crew-opencode
cp ~/.opencode/crew-opencode/config.json .opencode/crew-opencode/crew-opencode doctorThis will check:
- ✓ OpenCode directory exists
- ✓ Configuration is valid
- ✓ API keys are set
- ✓ Agents are configured
- ✓ SOPs are available
crew-opencode crew "Add a user login form with email and password fields"This will:
- PM analyzes the request and creates a plan
- TA researches best practices for login forms
- Design reviews UX patterns (if enabled)
- FE implements the login form component
- QA writes and runs tests
Expected Output:
🚀 Starting workflow: feature
📋 Total steps: 5
[1/5] PM - Planning
⚙ Analyzing request...
✓ Plan created
[2/5] TA - Research
⚙ Researching login form patterns...
✓ Technical specs ready
[3/5] FE - Implementation
⚙ Creating login component...
✓ Component implemented
[4/5] QA - Testing
⚙ Writing tests...
✓ Tests passing
[5/5] PM - Final Review
✓ Workflow complete!
📊 Summary:
Duration: 8m 32s
Tasks: 5 completed
Status: Success
crew-opencode crew "Fix the crash when clicking the logout button" --sop bugfixThis will:
- PM prioritizes the bug
- TA analyzes root cause
- FE implements minimal fix
- QA verifies fix doesn't break anything
crew-opencode crew "Refactor the authentication module to use hooks instead of class components" --sop refactorThis will:
- PM creates refactoring strategy
- TA analyzes dependencies
- FE performs refactoring
- QA ensures all tests still pass
SOPs (Standard Operating Procedures) define how the crew tackles different types of tasks.
1. Feature Development (feature)
- Duration: 60-90 minutes
- Steps: 6
- Agents: PM → TA → Design → FE → QA → PM
- Best for: New features, enhancements
2. Bug Fix (bugfix)
- Duration: 40-70 minutes
- Steps: 5
- Agents: PM → TA → FE → QA → PM
- Best for: Fixing bugs, small corrections
3. Refactoring (refactor)
- Duration: 100-160 minutes
- Steps: 6
- Agents: PM → TA → FE → QA → PM
- Best for: Code improvements, restructuring
# Default: feature SOP
crew-opencode crew "Add dark mode"
# Explicit SOP selection
crew-opencode crew "Fix memory leak" --sop bugfix
crew-opencode crew "Convert to TypeScript" --sop refactorYour crew consists of 5 specialized agents:
| Agent | Role | Model | Cost | Specialty |
|---|---|---|---|---|
| PM | Project Manager | Claude Opus 4.5 | 💰💰💰 | Strategy, orchestration |
| TA | Technical Analyst | Claude Sonnet 4.5 | 💰💰 | Research, architecture |
| FE | UI/UX Engineer | Gemini 3 Pro | 💰💰 | Implementation |
| Design | Designer | GPT 5.2 Medium | 💰💰 | UX/UI patterns |
| QA | Quality Assurance | Claude Haiku 4.5 | 💰 | Testing, verification |
crew-opencode listTo reduce costs, you can disable agents you don't need:
Edit your config file:
{
"crew": {
"design": {
"enabled": false // Disable Design agent
}
}
}Common configurations:
- Minimal (PM + TA + FE + QA): Best balance of quality and cost
- Full (All 5 agents): Maximum quality, higher cost
- Budget (PM + FE + QA): Lower cost, still functional
Preview what the crew will do without executing:
crew-opencode crew "Add user profiles" --dry-runRun crew-opencode in a different directory:
crew-opencode crew "Update homepage" --project /path/to/projectCheck the status of a running workflow:
crew-opencode status <workflow-id>When tasks fail, crew-opencode generates "Apology Letters" with:
- Root cause analysis
- Risk assessment
- Prevention strategy
View recent incidents:
crew-opencode reportsView specific incident:
crew-opencode reports --id <incident-id>View current configuration:
crew-opencode configGet specific value:
crew-opencode config crew.pm.modelSet configuration value:
crew-opencode config crew.pm.enabled trueSolution: Verify your API keys are set:
echo $ANTHROPIC_API_KEY
echo $OPENAI_API_KEY
echo $GOOGLE_API_KEYIf empty, add them to your shell profile and restart your terminal.
Solution: Run the install command to set up agents and SOPs:
crew-opencode install --globalSolution: Disable unnecessary agents or use simpler SOPs:
- Use
bugfixSOP for smaller changes - Disable Design agent if not needed
- Check if tasks can be broken into smaller chunks
Solution:
- Wait a few minutes and retry
- Upgrade your API plan for higher limits
- Reduce parallel agent execution
Solution: Review the incident report:
crew-opencode reportsThe QA agent will have documented:
- What went wrong
- Why it failed
- How to prevent it
Check system status:
crew-opencode doctorView logs:
ls -la .opencode/crew-opencode/workflows/Report issues:
- GitHub Issues: https://github.com/sehyun0518/crew-opencode/issues
- Documentation: https://github.com/sehyun0518/crew-opencode#readme
Now that you're set up:
- Try the examples above to get familiar with the crew
- Customize your configuration for your workflow
- Read the Configuration Guide for advanced options
- Review Example Workflows for inspiration
- Join the community to share your experience
Typical costs per workflow (approximate):
- Feature (60-90min): $0.50 - $2.00
- Bug Fix (40-70min): $0.30 - $1.00
- Refactor (100-160min): $1.00 - $3.00
Cost varies based on:
- Complexity of the task
- Which agents are enabled
- Your API plan rates
- Amount of context needed
Cost-saving tips:
- Disable Design agent for backend tasks
- Use bugfix SOP for simple changes
- Break large tasks into smaller workflows
- Keep context windows small
Need more help? Check out the full documentation or join discussions.