Skip to content

Agent

An Agent is an AI-powered entity that can:

  • Understand goals - Interprets your requests and breaks them into steps
  • Use tools - Invokes MCP servers and system tools to accomplish tasks
  • Iterate solutions - Refines answers through multiple rounds of thinking and action
  • Maintain context - Remembers conversation history across multiple exchanges
  • Long-term memory - Stores and retrieves important information across sessions

Unlike simple chat models, agents take autonomous actions to solve problems.

Core Concepts

Agent Loop

The execution pattern agents follow:

graph LR
    A[Receive Task] --> B[Analyze Situation]
    B --> C{Need Tools?}
    C -->|Yes| D[Call Tools]
    D --> E[Process Results]
    E --> B
    C -->|No| F[Return Answer]

Each cycle, the agent decides whether to respond directly or use tools to gather more information.

Configuration

Every agent has:

Component Purpose
Model The LLM powering reasoning (required)
Instruction Custom behavior and personality
Tools MCP servers + system tools available
Max Actions Iteration limit (1-200, default: 50)

Tool Access

Agents gain capabilities through tools:

  • MCP Servers - Service APIs, company database connections, code execution, web search...
  • System Tools - File operations, subagent, memory tools, system operations...

When to Use Agents

Use Case Example
Multi-step tasks "Analyze this codebase and create documentation"
Tool-based workflows "Search the web and summarize top 5 results"
Interactive problem solving "Debug this error by checking logs and code"
Autonomous execution "Monitor data and alert on anomalies"

Agent vs. Graph

Choose the right approach:

Feature Agent Graph (Workflow)
Structure Free-form, autonomous Predefined nodes and edges
Best for Open-ended tasks Structured processes
Control Model decides next steps You design the flow
Complexity Single reasoning loop Multiple specialized agents

Use Agents for flexibility. Use Graphs for reliability.

Next Steps