LLM Chat CLI

LLMChatCLI is a simple command-line tool for interacting with different LLM models and providers that use OpenAI-compatible API. It lets you load structured conversation inputs, engage in chat sessions, and automatically log responses for later analysis.

This tool is designed for testing and comparing outputs from different LLMs using consistent conversation history and prompts.

---

Features

• Chat with any OpenAI-compatible LLM provider via CLI.
• Load messages and system prompts from structured JSON or text files.
• Easily switch models or providers via flags or environment variables.
• Save complete chat logs automatically for future inspection.
• Supports ongoing conversations with pre-defined chat history.

---

Setup

1. Clone the Repository

git clone [email protected]:igorMSoares/llm-chat-cli.git
cd llm-chat-cli

2. Build the application

   This command will also automatically download and install the necessary dependencies.

   go build -o llm-chat-cli

   Go must be installed: Install Go

3. Create a .env file

   Copy the .env.example file to a new file named .env:

   cp .env.example .env

   Then, open the .env file and fill in the required environment variables:

   • LLM_PROVIDER_KEY: Your API key for the LLM provider.
   • LLM_MODEL: The name of the LLM model you want to use.
   • CHAT_COMPLETION_URL: The URL for the chat completion API (must be OpenAI compatible).
   • TEMPERATURE: The temperature for the LLM (optional, defaults to 0).

Usage

After building the application, you can run it from the terminal:

./llm-chat-cli [flags]

On Windows, the command would be:

.\llm-chat-cli.exe [flags]

Available Flags

The application supports the following command-line flags:

Flag	Description
--api-key	API key for the LLM provider (overrides LLM_PROVIDER_KEY)
--model	Name of the LLM model to use (overrides LLM_MODEL)
--url	Chat API URL (overrides CHAT_COMPLETION_URL)
--temperature	Sampling temperature (overrides TEMPERATURE)
--input	Input file name (default: messages.json)
--input-dir	Directory containing input files (default: input)
--prompts-dir	Directory containing prompt files (default: prompts)
--logs-dir	Directory where conversation logs will be saved (default: logs)

Example

To run using the example messages:

./llm-chat-cli --input messages.example.json

Input File

The input file is a JSON file that contains an array of messages, which can be used to set the context for the conversation or to load an ongoing chat history.

Each message is an object with the following properties:

• role: The role of the message sender. Can be user, assistant, or system.
• content: The content of the message.
• file: (Optional) The name of a file containing the system message. This is only used for system messages and will be loaded from the directory specified by --prompts-dir.

The input file can not be empty. It must be valid JSON and contain at least an empty array: [].

Behavior on Startup

The application's initial behavior depends on the role of the last message in the input file:

• If the last message is from the system or assistant, the application will prompt you for input to start the conversation.
• If the last message is from the user, the application will immediately send the entire conversation history to the LLM, display the assistant's response, and then prompt you for your next message.

Examples

1. Starting with a system prompt:

In this example, the system message is read from the prompts/system_prompt.md file. The application will start by asking for user input because the last message is not from the user.

[
  {
    "role": "system",
    "file": "system_prompt.md"
  }
]

2. Loading a conversation history:

This example loads a pre-existing conversation. Since the last message is from the user, the application will start by sending the request to the API to get the assistant's response before prompting for the next user input.

[
  {
    "role": "system",
    "content": "You are a helpful assistant."
  },
  {
    "role": "user",
    "content": "Hello, who are you?"
  },
  {
    "role": "assistant",
    "content": "I am a helpful assistant. How can I help you today?"
  },
  {
    "role": "user",
    "content": "What is the capital of Butan?"
  }
]

Commands

While chatting with the model, you can use the following commands:

Command	Description
/quit	Save the conversation log and exit
/quit!	Exit immediately without saving the conversation

Contributing

Contributions are welcome and encouraged!

Ways to contribute:

• Open issues for bugs, ideas, or questions

• Submit pull requests to fix, improve, or add new features

• Improve documentation or add examples

To get started:

1. Fork the repository

2. Create a feature branch (git checkout -b my-feature)

3. Make your changes

4. Open a pull request with a clear description

If you're unsure where to start, feel free to open an issue or ask!

---

License

This project is licensed under the MIT License. You are free to use, modify, and distribute it.

---

Show Your Support

If you find this project helpful, consider starring the repo to show your support!