AI Tools

Open source local AI voice assistant that runs entirely on your computer without cloud connectivity. Supports multiple open source language models for natural language understanding and generation. Includes text-to-speech for voice output. Works on Windows, Mac, and Linux. Completely private, your voice and conversations stay on your computer. Extensible with plugins. Perfect for those who want a personal AI assistant with total privacy and control.

Zero-code interface from Hugging Face for training and deploying machine learning models without writing code. Upload your data and the tool handles training, optimization, and deployment. Works with text classification, token classification, question answering, and other tasks. Automatically selects and trains models for your task. Includes deployment to Hugging Face Spaces. Perfect for non-technical users or teams wanting to move fast. Free tier available.

Open source framework specifically designed for fine-tuning language models with minimal hassle. Simplifies the complex process of customizing language models for your specific tasks. Supports multiple fine-tuning techniques including LoRA and full parameter tuning. Works with popular models and hardware configurations. Includes monitoring and evaluation tools. Active community with good documentation. Removes barriers to training custom models. Free and community-driven.

Minimalist machine learning framework written in Rust for building and running language models efficiently. Prioritizes simplicity and performance over flexibility. Compiles to WebAssembly for running in browsers. Excellent for creating lightweight, fast ML applications. Used for building production ML systems with performance guarantees. Still growing but showing promise for systems-level ML work. Open source from Hugging Face.

Optimized CUDA kernel for extremely fast inference on quantized language models. Provides significantly faster inference than standard implementations. Uses specialized GPU optimization for Llama and compatible models. Reduces memory requirements through efficient quantization support. Used in production systems where speed is critical. Excellent for batch processing or high-throughput inference. Open source and actively maintained.

Easy-to-use interface for running quantized language models locally with minimal setup. Provides a browser-based frontend so you can interact with models through your web browser. Includes a local API server for integration with other applications. Works with quantized models to run on modest hardware. Supports Llama, Mistral, and other popular open source models. Great for beginners who want to run LLMs without complex configuration. Free and open source.

Hackable implementation of open source language models designed for easy customization and fine-tuning. Build and train your own language models locally with simple Python code. Includes pre-built recipes for common tasks like fine-tuning and inference. Fully transparent code so you understand exactly what's happening. Works with popular models like LLama, Mistral, and others. Perfect for researchers and developers who want to customize models. Open source from Lightning AI.

Python bindings for llama.cpp that enable running large language models directly in Python applications. Makes it simple to add local LLM inference to Python projects without external services. Lightweight and efficient implementation. Works with all models supported by llama.cpp. Easy to integrate into existing Python applications and scripts. Popular choice for data scientists and Python developers. Free and well-maintained.

Lightweight C++ inference server for running Llama models on CPU and GPU.

Low-Rank Adaptation training technique for efficiently fine-tuning large language models by training minimal parameters. Dramatically reduces the number of trainable parameters from billions to thousands. Reduces memory requirements and speeds up training. Maintains nearly equivalent model quality compared to full fine-tuning. Reference implementation by Microsoft with broad adoption. Fundamental technique for making model customization accessible. Open source.

Framework for merging multiple language models into a single, improved model. Combines the strengths of different models to create a hybrid model. Useful for combining a general model with a specialized model. Enables model fusion techniques. Works with various open source models. Growing technique in the AI community. Reduces the need to train multiple separate models. Open source.

Compiler framework that compiles large language models to run on any hardware from phones to browsers to servers. Enables deploying LLMs to edge devices and browsers without server infrastructure. Supports iOS, Android, WebGPU, and many other platforms. Dramatically reduces latency by running models locally. Reduces costs by eliminating cloud API calls. Open source and backed by major tech companies. Revolutionary for bringing AI to the edge.

Open source implementation that lets language models execute code on your computer through natural language commands. Write instructions in English and the tool converts them to Python or shell commands and executes them. Enables language models to interact with your file system and run programs. Works locally for privacy and offline operation. Extends model capabilities far beyond text generation. Growing project with increasing capabilities.

API gateway that provides unified access to multiple language model providers through a single interface. Route API calls to OpenAI, Anthropic, Google, Mistral, and many other providers. Switch between models without changing application code. Includes features like caching to reduce costs. Provides monitoring and analytics across all providers. Great for teams evaluating multiple models or wanting flexibility. Supports both open source and proprietary models.

Parameter-Efficient Fine-Tuning library from Hugging Face that enables training large language models with minimal resources. Fine-tune massive models with a fraction of the memory and time required by full training. Supports techniques like LoRA and prefix tuning. Works with popular language models from Hugging Face. Reduces training time and costs significantly. Perfect for teams with limited compute budgets. Open source and widely used.

Distributed inference system for running large language models collaboratively by connecting multiple computers. Enables swarm hosting where many participants together host a single large model. Splits model layers across different computers to achieve distributed inference. Reduces hardware requirements per participant. Maintains model quality equivalent to centralized inference. Interesting alternative to traditional cloud-based inference. Open source and experimental.

Self-hosted AI coding assistant powered by open-source language models.

Framework that accelerates language model fine-tuning by 2 to 5 times compared to standard implementations. Dramatically reduces memory usage while maintaining model quality. Speeds up training significantly to save time and compute costs. Works with popular models like LLama, Mistral, and others. Easy drop-in replacement for standard fine-tuning code. Growing adoption in teams doing model customization. Open source and actively developed.

Fast and easy-to-use library for running language models and serving them with high throughput. Optimizes memory usage through efficient attention algorithm and KV cache optimization. Reduces serving costs by up to 10 times compared to traditional approaches. Supports popular models like LLama, Mistral, Qwen, and many others. Works with OpenAI-compatible API for easy integration. Used in production by many companies to serve LLMs efficiently. Free and open source.

Fast unified toolchain for JavaScript and TypeScript that combines linting and formatting in one tool. Replaces both Prettier and ESLint with better performance and lower configuration complexity. Written in Rust for blazing fast execution. Provides comprehensive code quality checks with zero configuration needed. Smaller installation size than Prettier plus ESLint. Compatible with existing ESLint configurations. Growing community adoption. Free and open source.