AI CLI Tools

A command-line interface for access to core AI services: speech-to-text, large language models, and text-to-speech. This toolkit offers advanced AI capabilities directly from their terminal.

Key Features

• Speech-to-Text (STT): Transcribe speech from your microphone to text.
• Large Language Model (LLM): Interact with powerful language models for various tasks.
• Text-to-Speech (TTS): Convert text to natural-sounding speech.

Highlights

• Remote Server Support: All AI services can be installed on remote servers for enhanced performance and flexibility.
• Modular Design: Each tool functions independently and can be combined for complex workflows.
• Real-time Processing: Supports real-time transcription and speech synthesis.
• Multiple LLM Providers: Switch between different LLM providers (LMStudio, Anthropic, OpenAI, Ollama).
• Customizable TTS: Options for different TTS models and voices, including RVC support.

Project Status

This project is in early alpha stage. Servers aren't stable yet, also this current implementation only offers a working prototype with opinionated, streamlined commands. While customization options are limited at this phase, the foundation is set for potential expansion into a comprehensive CLI Swiss Army knife for STT, LLM, and TTS applications.

Roadmap

• Enhance customization options for each tool
• Expand API parameter accessibility
• Improve documentation and usage examples
• Gather user feedback for prioritizing new features

Tool Descriptions

Speech to Text (STT)

The STT command transcribes speech from your microphone to text.

Usage:

• Basic: stt
• File output: stt > file.txt

Examples:

C:\Dev\Test\ai-cli-tools>stt
This is detected speech.

C:\Dev\Test\ai-cli-tools>stt > transcription.txt
[Transcription saved to file]

Large Language Model (LLM)

The LLM command sends queries to a Large Language Model (using LM Studio on our server).

Usage:

• Inline query: llm your question here
• Piped input: echo your question | llm
• Speech input: stt | llm
• File output: llm your question > answer.txt
• System message output: echo your query or question | llm your system message
• Provider and model selection: llm your question --provider [lmstudio|anthropic|openai|ollama] --model [model_name]

Examples:

C:\Dev\Test\ai-cli-tools>llm When was Barack Obama born?
Barack Obama, the 44th President of the United States, was born on August 4, 1961.

C:\Dev\Test\ai-cli-tools>echo List 5 reputable sources for information on climate change: | llm > sources.txt

C:\Dev\Test\ai-cli-tools>llm Tell me about AI --anthropic
[Response from Anthropic's Claude model]

C:\Dev\Test\ai-cli-tools>llm Explain quantum computing --openai --model gpt-4
[Response from OpenAI's GPT-4 model]

Text-to-Speech (TTS)

The TTS command synthesizes text into speech in real-time.

Usage:

• Inline text: tts text to speak
• Piped input: echo text to speak | tts
• Full pipeline: stt | llm | tts
• File output: tts Hello, world! > output.wav
• RVC toggle: tts Hello, world! --rvc

Examples:

C:\Dev\Test\ai-cli-tools>tts Let me hear this
Let me hear this

C:\Dev\Test\ai-cli-tools>echo hey there | llm answer short | tts
[Speaks the short answer]

C:\Dev\Test\ai-cli-tools>stt | llm translate to english, only return the translation | tts --rvc > translation.wav
[Transcribes speech, translates it to English, applies RVC, and saves to file]

Advanced Usage

You can combine these tools in various ways to create powerful workflows:

stt | llm --anthropic | tts --rvc
[Transcribes speech, processes it with Anthropic's LLM, and speaks the result using RVC]

echo Explain relativity | llm --openai --model gpt-4 | tts > explanation.wav
[Processes text with OpenAI's GPT-4, then speaks and saves the result]

stt | llm translate to english, only return the translation | tts --xtts-voice french_voice.wav
[Transcribes speech, translates it to English, and speaks the translation with a French voice]

Prerequisites

• Python 3.10.9
• CUDA 12.1

Installation

1. Clone the Repository

   git clone https://github.com/KoljaB/ai_cli_tools.git
   cd ai_cli_tools

2. Create a Virtual Environment

   For Windows:

   _create_venv.bat

3. Activate the Virtual Environment

   For Windows:

   _start_venv.bat

4. Install Dependencies

   Execute in that virtual environment:

   pip install -r requirements.txt

5. Upgrade PyTorch to Use GPU

   Execute in venv:

   pip install torch==2.3.1+cu121 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu121

   (Adjust the CUDA version according to your system.)

6. Install models for TTS and RVC

   Execute in venv or main environment:

   For Windows:

   _install_models.bat

7. Install CLI commands

   Execute this on your main python or on your clients environment, NOT in the virtual environment.

   For Windows:

   pip install -r requirements_client.txt
   _install_cli_commands.bat

8. Start the servers

   Execute this on your main python or on your clients environment, NOT in the venv.

   For Windows:

   _start_stt_server.bat
   _start_llm_server.bat
   _start_tts_server.bat

   or execute

   stt-server
   llm-server
   tts-server

Environment Setup

To use certain LLM providers, set the following environment variables on the server machine:

• For OpenAI: OPENAI_API_KEY
• For Anthropic: ANTHROPIC_API_KEY

Notes

• You can use CTRL+C to immediately abort any command (stt, llm, tts).
• The system will prompt to start a local server if it's not running when you try to use a command.
• Default models:
  • OpenAI: gpt-4o-mini
  • Anthropic: claude-3-sonnet-20240229
  • Ollama: llama3.1