Deploy Your First Agent Team in 5 Minutes This guide will walk you through creating a multi-agent team with Hive, complete with memory, RAG, and MCP integration.
Step 1: Initialize Hive # Initialize Hive workspace hive init # This creates: # - .hive/ directory # - agents.yaml (your agent configuration) # - .env (API keys) Edit agents.yaml to define your team:
version : "1.0" agents : - name : researcher role : "Research information and gather data" llm : claude model : "claude-3-5-sonnet-20241022" memory : true tools : - web-search - wikipedia - name : coder role : "Write and review code" llm : claude model : "claude-3-5-sonnet-20241022" memory : true tools : - bash - edit-file - read-file - name : analyst role : "Analyze data and create reports" llm : claude model : "claude-3-5-sonnet-20241022" memory : true tools : - python - data-viz memory : type : vector provider : qdrant persist : true path : ./.hive/memory mcp : enabled : true Step 3: Set API Keys Edit .env with your credentials:
# Required ANTHROPIC_API_KEY = sk-ant-your-key-here # Optional (based on your needs) OPENAI_API_KEY = sk-your-openai-key GOOGLE_API_KEY = your-google-key # Memory (if using Qdrant) QDRANT_URL = http://localhost:6333 Start with in-memory storage for quick testing! Just set provider: in-memory in agents.yaml.
Step 4: Start Hive # Start Hive with hot-reload enabled hive start --hot-reload # You should see: # 🐝 Hive v1.0.0 # ✓ Agents spawned: researcher, coder, analyst # ✓ Memory system initialized # ✓ Hot-reload enabled Step 5: Give Your Team a Task # Assign a task to your research agent hive task assign researcher "Research the latest trends in AI agents and summarize key findings" # Or use the team coordinator hive task create "Build a simple web scraper for tech news" --assign-to coder Real-World Example: Research & Development Workflow
Research Phase
hive task assign researcher \ "Research best practices for building REST APIs in Python" The researcher agent:
Searches the web
Reads documentation
Stores findings in memory
Development Phase
hive task assign coder \ "Build a REST API based on the research findings" \ --context "Use the research from the previous task" The coder agent:
Accesses research from memory
Writes implementation code
Creates tests
Analysis Phase
hive task assign analyst \ "Analyze the API implementation and create performance recommendations" The analyst agent:
Reviews the code
Analyzes architecture
Generates report
Common Agent Team Configurations Content Creation Team
Development Team
Data Analysis Team
agents : - name : researcher role : "Research topics and gather sources" tools : [ web-search , wikipedia ] - name : writer role : "Create engaging content" tools : [ write-file ] - name : editor role : "Review and improve content" tools : [ read-file , edit-file ] Using Memory and RAG Agents automatically remember context:
# Agent stores information hive task assign researcher "Research Python async best practices" # Later, another agent can access this knowledge hive task assign coder "Write async code using best practices" \ --use-memory # Query memory directly hive memory search "async best practices" Hot-Reload in Action Update agents.yaml while Hive is running:
agents : - name : researcher role : "Research information and gather data" llm : claude temperature : 0.7 # Added parameter tools : - web-search - wikipedia - arxiv # Added new tool Save the file - changes apply instantly without restart!
# Check updated configuration hive agents list # researcher now has arxiv tool available Multi-Agent Collaboration Have agents work together on complex tasks:
# Create a workflow hive workflow create product-launch.yaml product-launch.yaml:name : product-launch description : Complete product launch workflow steps : - agent : researcher task : "Research target market and competitors" output : market-research - agent : writer task : "Create product description based on research" input : market-research output : product-copy - agent : analyst task : "Analyze product positioning and suggest improvements" input : product-copy output : analysis-report parallel : - agent : coder task : "Build landing page" - agent : writer task : "Write blog announcement" Execute the workflow:
hive workflow run product-launch Pro Tips
Begin with 2-3 specialized agents, then add more as needed: # Start with agents : - name : assistant role : "General purpose helper" # Then expand to agents : - name : researcher - name : coder - name : writer - name : analyst
Enable memory for agents that need context: - name : researcher memory : true # Needs to remember findings - name : calculator memory : false # Pure computation, no context needed
Monitor Agent Performance
Next Steps
Advanced Workflows Learn to:
Create complex multi-step workflows
Use conditional logic
Handle errors and retries
Custom Tools Build your own tools:
MCP tool integration
Custom Python functions
External API connectors
Memory & RAG Deep dive into:
Vector databases
Knowledge graphs
Semantic search
Production Deployment Scale to production:
Docker deployment
Kubernetes orchestration
Monitoring and logging
Troubleshooting
Check API keys and connectivity: hive config validate hive agents test researcher
Verify memory backend is running: # If using Qdrant docker run -p 6333:6333 qdrant/qdrant # Or switch to in-memory # In agents.yaml: provider: in-memory
Ensure file watching is enabled: hive start --hot-reload --verbose Congratulations! 🎉 You’ve deployed your first AI agent team with Hive. Your agents are ready to collaborate on complex tasks!
Join the Community Share your agent configurations and learn from others
