---

cli_buddy

---

Call Any LLM from Your Terminal

• Lightweight and Fast
• Supports over 100+ LLMs (GPT, Claude, Cohere, LLama 3, Gemini, Mistral, DeekSeek, and many other LLMs including all Open Source LLMs)
• Use your own API key (OpenRouter)
• Local LLM Support using Ollama
• Complete privacy - not a single analytic data is collected
• Works on Linux, macOS, and Windows
• Can also view GUI similar to many AI chat app if you want

Demo

Features

1. Code Generation: Ask for code and directly generate (or overwrite) a file with it.
   • Also available: option to copy to clipboard, ask for explanation, or chat with the context.
2. Shell Command Execution: Ask for a shell command and directly execute it in your terminal.
   • Also available: option to copy to clipboard or ask for explanation.
3. Continuous Chat: Chat with AI continuously in your terminal.
4. Session Management: Save chat history in JSON format locally and load it when starting a new session using -s <session_id>.
   • By default, saving chat history is disabled.
5. GUI: Want to see GUI? Just run buddy open to open the GUI. // Currently Disabled due to some unexpected issues

Getting Started 🚀

Installation

1. Installing as a package

   • Simply run:

   dart pub global activate cli_buddy

2. Installing Dart SDK

   • Install Dart SDK following the official Install Dart SDK document.
   • Check if Dart sdk is installed by running dart --version
   • Run dart pub global activate cli_buddy

3. Using binary

   • Download binary for your OS in /release (currently I only have MacOS so tested on MacOS but technically should work on all OS)
   • If your system is Unix (MacOS, Linux) run sudo chmod +x <path/to/binary>
   • Test if it runs: ./<path/to/binary> --help
   • Set alias according to your system (e.g. alias buddy='./<path/to/binary>')

Get API Key from OpenRouter

OpenRouter provides unified LLM APIs at (almost) the same cost as the original LLM API provider. Note: OpenRouter provides some free AI models with limitations. You can try out for free using those models without payments Note: I am not affliated with OpenRouter and I do not receive any benefits from them by any means.

Add API Key

buddy set --or-api-key <your openrouter key>

Example buddy.config

{
  "api_provider": "openrouter",
  "localEndpoint": "localhost:43210",
  "save_session": true,
  "save_online": false,
  "max_messages": 20,
  "openrouter_default_model": "openai/gpt-4o",
  "ollama_default_model": null,
  "ollamaEndpoint": "localhost:11434",
  "openrouter_key": null,
  "temperature": 0.3,
  "max_tokens": null,
  "top_p": null,
  "top_k": null,
  "frequency_penalty": null,
  "presence_penalty": null,
  "repetition_penalty": null,
  "min_p": null,
  "top_a": null,
  "seed": null,
  "logit_bias": null,
  "logprobs": false,
  "top_logprobs": null,
  "response_format": null,
  "stop": null,
  "cmd_prompt": "If there is a lack of details, provide most logical solution.\nEnsure the output is a valid shell command.\nIf multiple steps required try to combine them together in one command.\nProvide only plain text without Markdown formatting.\nDo not provide markdown formatting such as ```",
  "explain_prompt": "Provide short and concise explanation of your previous response about command or code.\nProvide only plain text without Markdown formatting.\nDo not provide markdown formatting such as ```",
  "code_prompt": "Provide only code as output without any description.\nProvide only code in plain text format without Markdown formatting.\nDo not include symbols such as ``` or ```python.\nIf there is a lack of details, provide most logical solution.\nYou are not allowed to ask for more details.\nFor example if the prompt is \"Hello world Python\", you should return \"print('Hello world')\".\n",
  "chat_prompt": "You are a Flutter/Dart developer. Don't be verbose in your response but focus on solving problem or coding. As for state management, use hooks_flutter and riverpod_hook packages.",
  "validate_prompt": "Your job is to verify if the provided code by the previous AI assistant is valid.\nProvide concise response unless asked for more details."
}

Basic Usage

# Shell command
$ buddy shell how can I update homebrew

# If you want to use special characters, use quotes
$ buddy shell "how can I update homebrew?"

# Coding
$ buddy code "Generate a random number between 1 and 10 in Dart"

# Chat
$ buddy chat "Tell me about the SSH protocol"

# Show Graphical User Interface
$ buddy open

# Show CLI version
$ buddy --version

# Show usage help
$ buddy --help

Info Command

The info command allows you to search for AI models in OpenRouter. Below are the available options and flags:

• -q, --query: Search for AI models by name, provider, or model type (e.g., text, image).
• -o, --order: Specify the order in which the results should be displayed. Allowed values: name, context, prompt, completion, image.
• -f, --config: Display the current configuration file. If it does not exist, create a new one.
• -c, --credits: Display the credits available in OpenRouter.
• -l, --list: List all AI models available in OpenRouter.
• -p, --parameters: Query the parameters of a specific AI model.
• -s, --sessions: (-s list) List saved chat histories in the default session folder or (-s <session_id>) view a specific chat history.

Example Usage

# Search for AI models by name
$ buddy info -q "claude"

# Search for AI models by provider
$ buddy info -q "openai"

# List all available AI models
$ buddy info -l

# Display the current configuration
$ buddy info -f

# Display the credits used for your OpenRouter account
$ buddy info -c

# Query the parameters of a specific AI model
$ buddy info -p "model_id"

# List saved chat histories
$ buddy info -s list

# View a specific chat history by session ID
$ buddy info -s "session_id"

Markdown format

• As for 'code', 'suggest', 'explain' mode, by default system prompt will discourage using markdown format, will remove it in the final output (for 'code copy/shell run/copy').
• As for 'chat' mode, markdown is allowed, but to use it, you need to pass '-m --markdown' flag. However, if markdown flag is used, real time streaming message will be disabled. Also implementing markdown in console is hard. Feel free to make a merge request if you did it.

Roadmap

• Writing code to a file then executing
• Indexing specified path and file extension for Rag
• Support for RAG (Retrieval-Augmented Generation)
• Automated comments for all speficied file path/file extension
• Automated improving all specified file path/file extension
• Support for Google Vertex AI API

Contributions

Contributions are welcome! Feel free to open issues or submit pull requests.

---