ADK-TSLLM Agents
Use language models for reasoning, decision-making, and tool usage
LLM Agents are the most common agent type in ADK TypeScript. They use Large Language Models for reasoning, understanding natural language, making decisions, and interacting with tools.
Unlike deterministic workflow agents that follow predefined paths, LLM agents are dynamic. They interpret instructions and context, deciding how to proceed, which tools to use, or whether to transfer control to another agent.
Core Concepts
Agent Identity
Every LLM agent needs a clear identity:
- Name: Unique identifier for the agent (required)
- Description: Summary of capabilities for other agents to understand
- Model: The LLM that powers the agent's reasoning
Instructions
The instruction parameter shapes how your agent behaves:
- Core task or goal definition
- Personality or persona
- Behavioral constraints
- Tool usage guidance
- Output format requirements
Effective Instructions
Be clear and specific, use examples for complex tasks, and explicitly guide when and why to use tools.
Tools Integration
LLM agents can use tools to extend their capabilities beyond text generation:
- Function tools for calculations and processing
- Built-in tools for web search, file operations, etc.
- Agent tools for delegation to other agents
- Custom tools for specific integrations
Configuration Options
🎛️ Generation Control
Adjust temperature, output length, and safety settings
📊 Structured Data
Use input/output schemas for JSON-based communication
🧠 Context Management
Control conversation history and memory inclusion
⚙️ Advanced Features
Planning, code execution, and callback integration
When to Use LLM Agents
LLM agents are ideal for:
- Conversational interfaces - Chat bots, assistants, Q&A systems
- Tool coordination - Agents that need to use multiple tools intelligently
- Dynamic decision making - When responses depend on context and reasoning
- Natural language processing - Understanding and generating human language
Key Features
Dynamic Behavior
LLM agents adapt their responses based on context, instructions, and available tools.
Tool Integration
Seamlessly combine multiple tools and capabilities within a single agent.
Multi-Agent Coordination
Can delegate tasks to other agents or be used as tools by other agents.
Flexible Configuration
Extensive options for customizing behavior, output format, and capabilities.
Working with LLM Agents
Basic Setup
Start with a simple agent that has a clear purpose and model configuration.
Add Instructions
Provide clear, specific instructions that guide the agent's behavior.
Integrate Tools
Add relevant tools that extend the agent's capabilities.
Fine-tune Configuration
Adjust generation parameters, schemas, and advanced options as needed.
Related Topics
🛠️ Tools
Learn about available tools and creating custom ones
🧠 Models
Configure and choose between different LLM models
👥 Multi-Agent Systems
Coordinate multiple LLM agents for complex tasks
📞 Callbacks
Hook into agent execution for monitoring and control
Best Practices
Common Pitfalls
- Avoid vague instructions that lead to inconsistent behavior
- Don't overload agents with too many tools or responsibilities
- Test instructions thoroughly with different inputs
- Consider structured output for data processing tasks
Instruction Design
- Be specific about the agent's role and responsibilities
- Provide examples for complex output formats
- Clearly define when and how to use tools
- Set appropriate boundaries and constraints
Tool Selection
- Choose tools that match the agent's purpose
- Avoid redundant or conflicting tool capabilities
- Document tool usage in instructions
- Test tool interactions thoroughly
Performance Optimization
- Use appropriate models for your use case
- Configure generation parameters for consistency
- Implement proper error handling
- Monitor and log agent behavior