Planning Agents

Chain-of-thought reasoning with Plan → Act → Observe → Reflect loops.

Overview

Planning agents break down complex tasks into sequential steps, executing each one and adjusting the plan based on observations.

Key benefits:

  • Better handling of multi-step tasks
  • Explicit reasoning process
  • Self-correction during execution
  • Transparent decision-making

How It Works

  1. Plan: Agent creates step-by-step plan
  2. Act: Execute current step
  3. Observe: Analyze results
  4. Reflect: Adjust plan based on observations
  5. Repeat: Continue until goal achieved

Configuration

{
  planning: {
    enabled: true,
    max_steps: 10,
    enable_reflection: true,
    enable_tools: true,
    max_tool_calls: 20
  }
}

Usage

Enable planning mode:

curl -X POST http://localhost:8080/v1/chat/completions \
  -d '{
    "model": "phi3",
    "mode": "planning",
    "messages": [{"role": "user", "content": "Design a REST API for a blog"}]
  }'

Response Structure

{
  "choices": [{
    "message": {
      "content": "Final answer..."
    }
  }],
  "planning_result": {
    "goal": "Design REST API for blog",
    "steps": [
      {
        "step_number": 1,
        "thought": "Need to identify core entities",
        "action": "List entities: posts, users, comments",
        "observation": "Core entities identified",
        "reflection": "Good start, need authentication"
      }
    ],
    "total_steps": 5,
    "success": true
  }
}

Use Cases

  • System design: Architecture planning
  • Complex problem solving: Multi-step reasoning
  • Research tasks: Information gathering and synthesis
  • Debugging: Root cause analysis

With Tool Use

Combine planning with tool calling for powerful automation:

{
  planning: {
    enabled: true,
    enable_tools: true
  },
  tools: {
    enable_http: true,
    enable_filesystem: true
  }
}

The planner can now call APIs, read files, and execute shell commands as part of its plan.