CleverChatty CLi

The CLI application based on the CleverChatty package.

⚠️ UPDATE:
This project is now archived. Development has moved to the CleverChatty repository.
You can use the cleverchatty-cli package from there as well.

This application is the comand line AI chat tool. It allows to comunicate with veriety of LLMs and it has support of external tools calls using the MCP(Model Context Protocol).

Quick run

Install some model with ollama, for example, qwen2.5:3b:

ollama list
NAME                   ID              SIZE      MODIFIED    
.... 
qwen2.5:3b             357c53fb659c    1.9 GB    3 weeks ago 
....

And chat with this model.

go run github.com/gelembjuk/cleverchatty-cli@latest -m ollama:qwen2.5:3b

Or, when you clone the repo, you can run it with:

go build
./cleverchatty-cli -m ollama:qwen2.5:3b

However, this tool is really useful when you use some MCP servers. It is possible by creating a config file and adding the list of MCP servers to it.

Config

The config is not required. Most options from the config can be overriden with the command line arguments.

All supported models providers are:

• ollama
• google
• anthropic
• openai

For openai, anthropic, google it is needed to have additional arguments - API key and API endpoint in some cases. See the config file example below. It is possible to provide the additional arguments with the config file or with the command line options.

The config should be stored in config.json file.

{
    "log_file_path": "",
    "model": "ollama:qwen2.5:3b",
    "system_instruction": "Talk to a user like you are Elvis Presley.",
    "mcpServers": {
        "File_Storage_Server": {
            "url": "http://localhost:8000/sse",
            "headers": [
                "Authorization: Basic Q0F************MQ=="
            ]
        },
        "Some_STDIO_Server": {
            "command": "uv",
            "args": [
                "-m",
                "uvicorn",
                "example:app",
                "--host"
            ]
        }
    },
    "anthropic": {
        "apikey": "sk-**************AA",
        "base_url": "https://api.anthropic.com/v1",
        "default_model": "claude-2"
    },
    "openai": {
        "apikey": "sk-********0A",
        "base_url": "https://api.openai.com/v1",
        "default_model": "gpt-3.5-turbo"
    },
    "google": {
        "apikey": "AI***************z4",
        "default_model": "google-bert"
    }
}

The section mcpServers contains all MCP servers config. It supports both STDIO and SSE servers.

Usig the memory server

This tools allows to use long-term AI memory using a MCP server of specific interface.

"mcpServers": {
        "Memory_Server": {
            "url": "http://localhost:8001/sse",
            "headers": [
            ],
            "interface": "memory"
        },
        .... other MCP servers
    },

The value "interface": "memory" indicates that this server is a memory server. It will be used to store the chat history and the AI will be able to use it in the future conversations.

The interface requires a MCP server to have two tools - remember and recall.

Example:

@mcp.tool()
def remember(role: str, contents) -> str:
    """Remembers new data in the memory
    
    Args:
        role (str): The role of the data, e.g. "user", "assistant"
        contents (str): The contents to remember, usually the text of the message
    """

    Memory(config).remember(role, contents)

    return "ok"

@mcp.tool()
def recall(query: str = "") -> str:
    """Recall the memory"""
    
    r = Memory(config).recall(query)

    if not r:
        return "none"
    
    return r

remember tool accepts two arguments:

• role: The role of the data, e.g. "user", "assistant"
• contents: The contents to remember, usually the text of the message

recall tool accepts one argument:

• query: The query to search for the data in the memory. If empty, it is expected to return some common memories.

Using the Retrieval-Augmented Generation server

This tools allows to use RAG using a MCP server of specific interface.

"mcpServers": {
        "RAG_Server": {
            "url": "http://localhost:8002/sse",
            "headers": [
            ],
            "interface": "rag"
        },
        .... other MCP servers
    },

The value "interface": "rag" indicates that this server is a RAG server. It will be used to store the chat history and the AI will be able to use it in the future conversations.

The expected interface is the one tool named knowledge_search with two arguments: query and num.

Example:

func createServer() *server.MCPServer {
	// Create MCP server
	s := server.NewMCPServer(
		"Retrieval-Augmented Generation Server. Adding the company data to the AI chat.",
		"1.0.0",
	)

	execTool := mcp.NewTool("knowledge_search",
		mcp.WithDescription("Search for documents and information over the company data storage"),
		mcp.WithString("query",
			mcp.Required(),
			mcp.Description("The query to search for documents and information"),
		),
		mcp.WithNumber("num",
			mcp.Description("The number of results to return"),
		),
	)

	s.AddTool(execTool, cmdKnowledgeSearch)
	return s
}

To run the chat using this config stored in teh config.json, you can use the command:

 go run github.com/gelembjuk/cleverchatty-cli@latest --config config.json

Installation

Prerequisites

• Go 1.20 or later
• Ollama installed and running

Install

go install github.com/gelembjuk/cleverchatty-cli@latest