AI Agents & Automation

Llama-X is an open academic research project dedicated to advancing the performance of LLaMA models to state-of-the-art (SOTA) LLM capabilities. The project emphasizes a long-term, systematic, and rigorous approach, encouraging open-source community contributions. It aims to publish all code, models, data, and experimental details, continuously improving model versions and summarizing methods in academic papers. Llama-X focuses on key research areas such as instruction tuning, RLHF & RLAIF, data quality, long context transformers, multi-modal modeling, multilingual performance, efficient infrastructure, comprehensive evaluation, interpretability, and LLM on actions. The project provides a complete research plan and welcomes contributors to collaborate on iterative improvements, with new models requiring significant performance gains on automatic evaluations.

markpdfdown is a powerful open-source tool designed to simplify the conversion of PDF documents and images into clean, editable Markdown text. Leveraging advanced multimodal AI models through LiteLLM, it accurately extracts text and preserves formatting, including complex structures like tables, formulas, and diagrams. Key features include PDF to Markdown and Image to Markdown conversion, multi-provider support for OpenAI and OpenRouter, and a flexible command-line interface. It also offers a desktop application for a more user-friendly experience. The modular architecture ensures a clean and maintainable codebase, making it an ideal solution for developers and users needing precise document transcription.

muscle-mem is a Python SDK designed to act as a behavior cache for AI agents. It records an agent's tool-calling patterns as it solves tasks, and then deterministically replays these learned trajectories when the same task is encountered again. This approach aims to get Large Language Models (LLMs) out of the hotpath for repetitive tasks, significantly increasing speed, reducing variability, and eliminating token costs. The SDK allows for instrumenting tool functions and methods with decorators, and features a robust cache validation system using 'Checks' to ensure safe tool reuse. It also supports parameterization for dynamic arguments, making it adaptable to varying task inputs.

MMdnn is a comprehensive, open-source tool designed to simplify the interoperability of deep learning models across various frameworks. It provides essential functionalities such as model conversion, allowing users to train a model in one framework and deploy it in another. The tool also supports model visualization, offering an intuitive way to display network architectures. Additionally, MMdnn assists with model retraining by generating code snippets and provides guidelines for deploying deep learning models to different hardware platforms. It supports a wide range of popular frameworks including Caffe, Keras, MXNet, TensorFlow, CNTK, PyTorch, ONNX, and CoreML, making it a versatile solution for developers and researchers working with diverse deep learning ecosystems.

Elora provides generative AI chat and call assistants designed to automate and enhance business communications. It offers both internal chat assistants for streamlining information within companies and external chat assistants that integrate into websites to engage users. Additionally, Elora features incoming and outgoing call assistants to revolutionize the handling of repetitive calls, such as customer inquiries or follow-ups on unpaid invoices. The platform is designed for easy setup, requiring no coding, and allows users to monitor and optimize assistant performance from a central dashboard. Elora aims to improve customer satisfaction, boost productivity, and integrate seamlessly into existing business operations.

VibeVoice-Large is an AI-powered tool designed for creating podcast audio files. Users can input a script and then select distinct voice samples for different speakers, enabling the generation of dynamic and multi-voice podcast content. The application provides flexibility in specifying the number of speakers, making it suitable for various podcast formats, from interviews to narrative storytelling. This tool simplifies the audio production process by automating voice generation based on provided text and chosen vocal characteristics.

RAGEN (Reasoning AGENT) is a flexible reinforcement learning framework designed for training reasoning agents, particularly Large Language Models (LLMs), in interactive and stochastic environments. It introduces StarPO (State-Thinking-Actions-Reward Policy Optimization), a unified RL framework that supports multi-turn, trajectory-level agent training with fine-grained control over reasoning processes, reward assignment, and prompt-rollout structures. RAGEN-2, the latest iteration, includes SNR-Adaptive Filtering to mitigate noisy gradient updates and reasoning collapse diagnostics to detect and monitor template collapse during training. The framework is compatible with Gym environments and offers 10 built-in environments for diverse testing. It's ideal for researchers and developers focused on advancing the capabilities and stability of LLM-based agents.

Replica Studios was an AI voice platform that provided tools for text-to-speech and audio editing, catering to various creative projects including gaming and film production. The platform aimed to offer a user-friendly interface with styling and interactive elements for voice creation. However, Replica Studios has officially announced its closure, stating that it has signed off and is no longer operational. The company expressed gratitude to its users for their support during its journey.

SwitchAI is an open-source Android application designed to simplify the management of AI digital assistants on your device. It offers a fresh and streamlined approach, allowing users to easily select, start, and manage their preferred AI assistants. With SwitchAI, you can seamlessly switch between installed digital assistant apps, choose an assistant each time you activate your device's digital assistant feature, or set a default. It also supports quick access via home screen widgets and Quick Settings tiles. The tool boasts broad compatibility with a growing list of popular AI assistant apps, replacing older solutions like Plugin-VoiceGPT, and is ideal for anyone looking to optimize their interaction with multiple AI assistants on Android.

web-gpu-doc-chat is an innovative web application that brings a powerful Vicuna-7B language model directly into your browser environment. Users can upload or provide any text and then engage in conversational AI, asking questions or providing prompts, with the model responding instantly. This eliminates the need for server-side processing, offering a private and efficient way to interact with an LLM for document understanding and discussion. It's ideal for those who want to leverage advanced AI capabilities without external dependencies or data transfer.

simple_GRPO is an open-source implementation of the GRPO algorithm, specifically designed for reproducing r1-like LLM thinking. It utilizes a core loss calculation formula referenced from Hugging Face's trl, but with a significantly simplified codebase. The tool aims to save GPU memory, enabling feasible and efficient training, and helps users quickly understand and experiment with Reinforcement Learning processes like GRPO. It supports features such as improved multi-answer generation, regrouping, penalty on KL, and parameter tuning, all within approximately 200 lines of code across two files. The reference model is decoupled, allowing it to run on separate GPUs, which prevents multiple copies from being created by torch’s multiprocessing and enables training of large models on less powerful hardware.

Stable Diffusion WebUI Forge is an open-source platform that enhances the capabilities of Stable Diffusion WebUI, focusing on improving development workflows, optimizing resource management, and accelerating inference speeds. Inspired by 'Minecraft Forge,' it aims to become the definitive 'Forge' for SD WebUI. The platform is currently based on SD-WebUI 1.10.1 and synchronizes with the original WebUI periodically. It offers features like GPU memory management, support for various LoRAs, preprocessors, ControlNets, and IP-Adapters. Forge also integrates Gradio 4 UIs and provides one-click installation packages for different CUDA/Pytorch versions, making it accessible for users to quickly set up and run the environment.

WhiStress Demo is an AI-powered tool available on Hugging Face that provides audio transcription with a unique feature: it highlights emphasized words. Users can easily interact with the tool by either uploading an audio file or recording their voice directly within the interface. The platform is designed to offer clear transcriptions, with a recommendation to speak clearly for optimal results. This tool is particularly useful for analyzing speech patterns and identifying key stress points in spoken language, making it valuable for various applications from linguistic analysis to speech therapy demonstrations.

WideLabs specializes in delivering sovereign AI infrastructure tailored for businesses. The platform provides robust GPU cloud services, enabling companies to run demanding AI workloads efficiently. Beyond infrastructure, WideLabs also develops and integrates proprietary AI models, offering advanced capabilities for various business needs. Their end-to-end solutions ensure comprehensive support from deployment to ongoing management, addressing complex challenges in generative AI, computer vision, and predictive algorithms. WideLabs aims to create a significant impact on individuals, institutions, and companies by leveraging cutting-edge AI technologies.

🔬🧠GPT4O🖼️🎥 is a versatile AI chatbot available on Hugging Face that leverages GPT-4 to process and analyze various forms of input, including text, audio, images, and video files. Users can upload these different modalities and receive detailed responses, summaries, and analyses. Beyond its multimodal AI capabilities, the tool also integrates a feature for searching scholarly articles on ArXiv, making it useful for research and information retrieval. This application is designed to provide comprehensive insights across diverse data types.

Unsloth is an open-source platform designed for training and running a wide array of open models, including Gemma 4, Qwen3.5, DeepSeek, and gpt-oss, directly on local machines. It offers a user-friendly web UI, Unsloth Studio, for easy interaction, alongside a code-based version, Unsloth Core. The tool boasts significant performance improvements, enabling up to 2x faster training with up to 70% less VRAM, without compromising accuracy. It supports various model types including text, audio, embedding, and vision models, and provides features like model inference, export, tool calling, and code execution. Unsloth also includes advanced training capabilities such as reinforcement learning, custom Triton kernels, and data recipes for dataset creation from diverse file types.

UltraRAG is a lightweight RAG development framework based on the Model Context Protocol (MCP) architecture, designed for both research exploration and industrial prototyping. It standardizes core RAG components like Retriever and Generation as independent MCP Servers, allowing for precise orchestration of complex control structures such as conditional branches and loops through simple YAML configuration. The platform features a visual RAG Integrated Development Environment (IDE) with a Pipeline Builder that supports bidirectional real-time synchronization between canvas construction and code editing. This enables granular online adjustments of pipeline parameters and prompts, along with an Intelligent AI Assistant for structural design, parameter tuning, and prompt generation. UltraRAG aims to lower the barrier to entry for building RAG systems and accelerate deployment, offering one-click conversion of logic flows into interactive dialogue systems and integrated knowledge base management.

MarvelAI is a revolutionary AI-first business intelligence platform designed to transform businesses by integrating artificial intelligence into every process, decision, and strategy. It leverages a cloud-native architecture for instant global deployment, unlimited scalability, and agility, allowing businesses to adapt faster than market changes. The platform focuses on revolutionizing customer interactions, creating emotional connections, predictive services, and exceptional experiences that become a competitive advantage. MarvelAI aims to deliver significant business impact, including explosive revenue growth, impactful efficiency gains, and accelerated decision-making through real-time insights. It serves various industries such as Banking & Financial Services, Insurance, Retail, Healthcare, Manufacturing, and Technology, offering tailored AI solutions for each sector.

Gradio Llama2.mojo provides a platform for users to interact with and experiment with the Llama2 language model. Hosted on Hugging Face Spaces, this tool is built with Gradio and utilizes a Docker SDK, making it accessible for those interested in exploring large language models. While the live website currently indicates a runtime error, the intention of the tool is to offer a free environment for engaging with Llama2. It serves as a valuable resource for developers, researchers, and AI enthusiasts looking to understand the capabilities and behaviors of this specific AI model.

OpenManus-RL is an open-source initiative, collaboratively led by Ulab-UIUC and MetaGPT, dedicated to advancing reinforcement learning (RL) tuning for large language model (LLM) agents. Inspired by successful RL tuning in models like Deepseek-R1, this project explores novel algorithmic structures, diverse reasoning paradigms, and sophisticated reward strategies. It supports rigorous testing on agent benchmarks such as GAIA, AgentBench, WebShop, and OSWorld, with all progress and tuned models openly shared. The platform integrates advanced RL algorithms like PPO and DPO through the Verl submodule, offering efficient and flexible training capabilities. It also provides a simplified library for Supervised Fine-Tuning (SFT) and GRPO tuning, making it a comprehensive solution for researchers and developers looking to push the boundaries of agent reasoning and tool integration.

Intics provides Agentic Document Intelligence (ADI) to revolutionize document processing by handling 100% of documents, including complex unstructured or handwritten ones. Unlike traditional methods, Intics offers a no-touch ADI system with full autopilot feedback loops, ensuring high accuracy and efficiency. It leverages pre-trained large vision models (Krypton and Radon) for data extraction without the need for additional training. The platform is designed to work across various industries and data types, offering a scalable solution for processing millions of documents. Intics aims to eliminate manual intervention, reduce costs, and provide real-time control over the data extraction process, transforming dormant document assets into actionable intelligence for autonomous enterprises.

ToDoIt is an innovative voice and AI-powered to-do list application designed to help users manage tasks efficiently. By simply speaking their daily goals, users can create tasks in less than 10 seconds, allowing them to focus on execution rather than manual entry. The tool supports 57 languages for voice transcription and offers AI-powered task recommendations to enhance productivity. It prioritizes user privacy by encrypting task titles and instantly deleting audio files after transcription. ToDoIt is currently available as a web version, fully responsive across all devices, with mobile apps planned for future development.

ai-agents-masterclass is a comprehensive GitHub repository designed to accompany an AI Agents Masterclass video series. It offers all the code and resources used in the YouTube series, enabling developers to follow along and build their own AI agents. The masterclass focuses on empowering developers to leverage AI agents for transforming businesses and creating sophisticated software. The repository includes examples for building agents with LangChain, LangGraph, n8n, and other technologies, covering topics from basic agent creation to RAG agents, task management, and deployment. It serves as a practical guide for anyone looking to dive deep into AI agent development.

AgentGym-RL is a comprehensive framework designed for training Large Language Model (LLM) agents to excel in long-horizon, multi-turn interactive decision-making tasks using reinforcement learning. It addresses challenges in existing methods by offering a modular system that supports a wide array of real-world scenarios and integrates mainstream RL algorithms. The framework introduces ScalingInter-RL, a progressive horizon-scaling strategy that balances exploration and exploitation, leading to stable and efficient optimization. It includes diverse environments like Web Navigation, Deep Search, Digital Games, Embodied Tasks, and Scientific Tasks, and supports various training paradigms beyond online RL, such as SFT, DPO, and AgentEvol. AgentGym-RL also provides a visualized interactive user interface for analyzing interaction trajectories.