sokrates 0.3.0

A collection of tools for LLM interactions like prompt refinement and idea generation. It also provides a console chat client with speech to text feature.

sokrates

A collection of tools for LLM interactions and system monitoring, designed to facilitate working with Large Language Models (LLMs) through modular components, well-documented APIs, and production-ready utilities.

Table of Contents

• Description
• Installation
• Usage
  • Available Commands
• Features
• Contributing
• License
• Contact
• Changelog

Description

sokrates is a comprehensive framework for working with Large Language Models (LLMs). It provides:

• Advanced prompt refinement tools
• System monitoring for resource tracking during LLM operations
• An extensive CLI interface for rapid experimentation
• Modular components and well-documented APIs

The project includes utilities for:

• Managing configuration settings
• Interacting with OpenAI-compatible LLM APIs
• Processing and cleaning LLM-generated text
• Monitoring system resources in real-time
• Executing complex workflows for idea generation and prompt refinement

Installation via pip

pip install sokrates

# or via uv
uv pip install sokrates

# test the installation
list-models --api-endpoint http://localhost:1234/v1

# or via uv
uv run list-models --api-endpoint http://localhost:1234/v1

Installation for development purposes

Prerequisites: Python 3.9 or higher

git clone https://github.com/Kubementat/sokrates.git
cd sokrates
uv sync

Configuration

You can configure the library via a configuration file in $HOME/.sokrates/.env

# Copy
cp .env.example $HOME/.sokrates/.env

# adjust to your needs
vim $HOME/.sokrates/.env

Whisper dependencies

brew install ffmpeg
brew install whisper-cpp

Usage

Basic Command Structure

Most commands follow this structure:

command --option1 value1 --option2 value2

You can always display the help via:

command --help

e.g.

uv run list-models --help

Available Commands

• benchmark-model: Benchmark LLM models
• benchmark-results-merger: Merge benchmark results
• benchmark-results-to-markdown: Convert benchmark results to markdown
• execute-tasks: Executes a list of tasks sequentially from a given task list in json format (generated by breakdown_task)
• fetch-to-md: Fetch content and convert to markdown
• generate-mantra: Generate mantras or affirmations
• list-models: List available LLM models
• idea-generator: Generate ideas using a multi-stage workflow
• refine-and-send-prompt: Refine and send prompts to a LLM
• refine-prompt: Refine prompts for better LLM performance
• breakdown-task: Break down complex tasks into manageable steps
• send-prompt: Send a prompt to a LLM API
• llmchat: Chat with LLMs via the command line

Example Usage

# Getting help for a command and usage instructions
uv run refine-prompt --help

# List available models
uv run list-models --api-endpoint http://localhost:1234/v1

# Generate ideas
uv run idea-generator --output-directory tmp/ideas --verbose

# Benchmark a model
uv run benchmark-model --model qwen/qwen3-8b --iterations 5

Features

• Prompt Refinement: Optimize LLM input/output with advanced refinement tools
• System Monitoring: Real-time tracking of resource usage during LLM operations
• CLI Interface: Extensive command-line tools for rapid experimentation
• Modular Architecture: Easily extendable and customizable components
• Testing Infrastructure: Built-in test framework with pytest integration

Contributing

1. Fork the repository and create a new branch
2. Make your changes and add tests if necessary
3. Submit a pull request with a clear description of your changes

Please read CONTRIBUTING.md for more details.

License

This project is licensed under the MIT License. See LICENSE for details.

Contact

• [e-mail]
• GitHub: @julweber
• Linked.in : Julian Weber

Changelog

View our CHANGELOG.md for a detailed changelog.