70-server-agent-cli-stream

README

Server Agent in CLI Mode with StreamCompletion

This example demonstrates how to use a Server Agent in CLI mode using the StreamCompletion method, which provides the same functionality as the Crew Agent's streaming method.

Key Features

• Dual Mode: The same agent can be used in both CLI and HTTP server modes
• StreamCompletion: Uses the crew-style StreamCompletion method for CLI interactions
• Tool Execution: Supports tool calls with automatic confirmation in CLI mode
• Streaming: Real-time streaming responses
• Consistent API: Same interface as Crew Agent for easy code migration

What's New

The Server Agent now has a StreamCompletion method that mirrors the Crew Agent's functionality:

agent.StreamCompletion(question, func(chunk string, finishReason string) error {
    if chunk != "" {
        fmt.Print(chunk)
    }
    return nil
})

This method handles:

1. Context compression (if configured)
2. Tool call detection and execution
3. RAG context integration (if configured)
4. Streaming response generation

CLI vs Server Mode

CLI Mode (this example)

• Uses StreamCompletion method
• Interactive console-based interface
• Auto-confirms tool execution by default
• Can set custom confirmation prompts with SetConfirmationPromptFunction

Server Mode (sample 49-server-agent-stream)

• Uses StartServer method
• HTTP API with SSE streaming
• Web-based confirmation via /operation/validate endpoint
• Separate endpoints for all operations

Usage

Prerequisites

Make sure you have a model server running:

# Using Docker Model Runner or Ollama
docker model pull ai/qwen2.5:1.5B-F16
# or
docker model pull hf.co/menlo/jan-nano-gguf:q4_k_m

Run the Example

cd samples/72-server-agent-cli-stream
go mod tidy
go run main.go

Try These Commands

🧑 You: Hello! Say hello to Alice
🤖 Bob: [executes say_hello tool and responds]

🧑 You: What is 42 plus 58?
🤖 Bob: [executes calculate_sum tool and responds]

🧑 You: What time is it?
🤖 Bob: [executes get_current_time tool and responds]

Customization

Custom Confirmation Prompt

By default, tool calls are auto-confirmed in CLI mode. To add manual confirmation:

agent.SetConfirmationPromptFunction(func(functionName string, arguments string) tools.ConfirmationResponse {
    fmt.Printf("\n⚠️  Execute %s? (y/n): ", functionName)
    response, _ := input.ReadInput()
    if response == "y" {
        return tools.Confirmed
    }
    return tools.Denied
})

Add RAG Support

ragAgent, _ := rag.NewAgent(ctx, ragConfig, modelConfig, embeddingConfig)
agent.SetRagAgent(ragAgent)
agent.SetSimilarityLimit(0.7)
agent.SetMaxSimilarities(3)

Add Context Compression

compressorAgent, _ := compressor.NewAgent(ctx, compressorConfig, modelConfig)
agent.SetCompressorAgent(compressorAgent)
agent.SetContextSizeLimit(8000)

Benefits of Using Server Agent in CLI Mode

1. Code Reusability: Same agent code works in both CLI and HTTP modes
2. Testing: Test your server agent logic in CLI before deploying as HTTP service
3. Consistency: Same behavior across crew agents and server agents
4. Flexibility: Easy to switch between interactive and server modes

Related Examples

• 49-server-agent-stream: Server agent in HTTP mode
• 50-server-agent-with-tools: Server agent with tools (HTTP mode)
• 54-server-agent-tools-rag-compress: Full-featured server agent (HTTP mode)
• 55-crew-agent: Crew agent with StreamCompletion
• 69-simple-crew-agent: Simple crew agent example

Architecture

ServerAgent
  ├── CLI Mode (this example)
  │   └── StreamCompletion()
  │       ├── Compress context if needed
  │       ├── Detect and execute tool calls
  │       ├── Add RAG context
  │       └── Stream response
  │
  └── HTTP Mode (sample 49)
      └── StartServer()
          └── handleCompletion()
              ├── Compress context if needed
              ├── Detect and execute tool calls
              ├── Add RAG context
              └── Stream response via SSE

Both modes share the same underlying logic but differ in:

• Input/output handling
• Tool confirmation mechanism
• Response streaming method