multi-agent-dialog

Multi-Agent Dialog Example

An interactive multi-agent dialog system demonstrating agent communication, tool usage with ACE (Agent Capability Engine), and conversation history tracking.

Tech Stack and Components

Technologies Used

• Language: Go 1.26+
• LLM Provider: Ollama (llama3.2) or other OpenAI API-compatible services
• Configuration Format: YAML
• ACE Engine: Agent Capability Engine for tool filtering
• Concurrency Control: errgroup
• Interactive Interface: Command Line (CLI)

Core Components Used

Supported Tools and Capabilities

Key Feature Implementations

Code Location References:

• Agent creation and initialization: examples/multi-agent-dialog/main.go:50-80
• ACE tool filtering: examples/multi-agent-dialog/main.go:120-150
• Tool call handling: examples/multi-agent-dialog/main.go:200-250
• Multi-turn dialog management: examples/multi-agent-dialog/main.go:150-180
• Capability list display: examples/multi-agent-dialog/main.go:280-310

Features

ACE (Agent Capability Engine)

Intelligent tool filtering system that dynamically selects relevant tools based on user queries:

• Detect Capabilities: Identify needed capability types (math, time, network, etc.) from query
• Filter Tools: Show LLM only 2-4 relevant tools instead of all tools
• Improve Accuracy: Reduce tool selection errors
• Lower Cost: Reduce token usage

Code Location: examples/multi-agent-dialog/main.go:120-150

Multi-Turn Dialog

Supports up to 3 rounds of tool call loops:

User Input → LLM Analyzes → Tool Call → Tool Result → LLM Generates → Answer
         ↑                                              │
         └──────────────── Add History ──────────────────┘

Code Location: examples/multi-agent-dialog/main.go:150-180

Capability Query

Support natural language queries about agent capabilities:

User: What can you do?
Agent: I can help you with:
  - Math calculations (calculator)
  - Time queries (datetime)
  - HTTP requests (http_request)
  ...

Code Location: examples/multi-agent-dialog/main.go:280-310

Quick Start

1. Configure LLM

Edit config/server.yaml:

llm:
  provider: "ollama"
  base_url: "http://localhost:11434"
  model: "llama3.2"
  timeout: 60
  max_tokens: 2048

2. Run the Example

cd examples/multi-agent-dialog
go run main.go

3. Interactive Example

=== Multi-Agent Dialog System ===
Available commands:
  - Type your question
  - 'capabilities' - Show all available tools
  - 'exit' - Quit

> What can you do?
I can help you with:
  - Math calculations (calculator)
  - Time queries (datetime)
  - HTTP requests (http_request)
  - Knowledge retrieval (knowledge_search)
  - File operations (file_read)
  - Memory operations (memory_read)

> Calculate sum from 1 to 100
[TOOL:calculator {"expression":"1+2+...+100"}]
The result is: 5050

> What time is it?
[TOOL:datetime {"operation":"current_time"}]
Current time: 2026-03-23 14:30:45

> Request Baidu homepage
[TOOL:http_request {"url":"https://www.baidu.com"}]
HTTP request successful, status code: 200
Response size: 15,234 bytes

Configuration

config/server.yaml

llm:
  provider: "ollama"
  base_url: "http://localhost:11434"
  model: "llama3.2"
  timeout: 60
  max_tokens: 2048

tools:
  enabled: true
  capabilities:
    - math
    - time
    - network
    - knowledge
    - file
    - memory

Architecture

User Input
    │
    ▼
┌─────────────────────┐
│  Capability Query  │
│  Detection          │
└────────┬────────────┘
         │
    ┌────┴────┐
    │ Yes     │ No
    ▼         ▼
┌────────┐ ┌─────────────────────┐
│Show    │ │ ACE Capability      │
│List    │ │ Detection           │
└────────┘ │ - Analyze query     │
           │ - Match capability   │
           └────────┬────────────┘
                    │
                    ▼
           ┌─────────────────────┐
           │  LLM Generate       │
           │  Response           │
           │  - Use tool schemas │
           │  - Generate calls   │
           └────────┬────────────┘
                    │
                    ▼
           ┌─────────────────────┐
           │  Tool Call Handling │
           │  - Parse [TOOL:]    │
           │  - Execute tools    │
           │  - Return results   │
           └────────┬────────────┘
                    │
                    ▼
           ┌─────────────────────┐
           │  Generate Final     │
           │  Answer             │
           └─────────────────────┘

ACE Workflow

User Query: "Calculate 1+1 and tell me the time"
    │
    ▼
Capability Detection: [math, time]
    │
    ▼
Tool Filtering: [calculator, datetime]
    │
    ▼
LLM Prompt: "Available tools: calculator, datetime..."
    │
    ▼
Tool Calls: [TOOL:calculator, TOOL:datetime]
    │
    ▼
Result Integration: "1+1=2, current time is 14:30"

Code Location: examples/multi-agent-dialog/main.go:120-250

Extending

Add New Tool

1. Implement tool interface:

type MyTool struct{}

func (t *MyTool) Name() string {
    return "my_tool"
}

func (t *MyTool) Execute(ctx, params) (Result, error) {
    // Implementation
}

1. Register tool:

agentTools.RegisterTool(&MyTool{})

Customize Capability Mapping

Modify capability keywords in examples/multi-agent-dialog/main.go:

capabilities := map[string]string{
    "math": "calculate,sum,multiply,divide",
    "time": "time,date,current time",
    // Add more
}

Troubleshooting

Issue 1: Tool Not Recognized

No tools available for query

Solution:

• Check if query contains capability keywords
• View capability detection results in logs
• Try more explicit query

Issue 2: Tool Call Failed

Error: tool execution failed

Solution:

• Check if tool parameters are correct
• View detailed error logs for tool execution
• Verify tool-dependent services are available

Issue 3: LLM Not Calling Tools

LLM response doesn't contain [TOOL:] marker

Solution:

• Check if LLM model supports tool calls
• Adjust prompt template
• Try different LLM model

References

• Main README
• ACE Documentation
• Tools System Documentation
• LLM Documentation

License

MIT License

Created: 2026-03-23
Example Type: Multi-Agent Dialog Demonstration
Code Location: examples/multi-agent-dialog/main.go:1-358