AI Agents & Automation

FunASR is a fundamental end-to-end speech recognition toolkit designed to bridge the gap between academic research and industrial applications. It offers a comprehensive suite of features including speech recognition (ASR), Voice Activity Detection (VAD), Punctuation Restoration, Language Models, Speaker Verification, Speaker Diarization, and multi-talker ASR. The toolkit provides convenient scripts and tutorials for both inference and fine-tuning of pre-trained models. FunASR boasts a vast collection of academic and industrial pre-trained models available on ModelScope and Hugging Face, including the highly accurate and efficient Paraformer-large. Recent updates include support for large models like Fun-ASR-Nano-2512 (31 languages), Whisper-large-v3-turbo, and Qwen-Audio multimodal models, alongside continuous improvements in real-time and offline transcription services, memory optimization, and multi-platform support.

Magentic-UI is a research prototype of a human-centered AI agent designed to automate complex web and coding tasks that may require monitoring. Unlike black-box agents, the system reveals its plan before executions, lets users guide its actions, and requests approval for sensitive operations while browsing websites, executing code, and analyzing files. Key features include co-planning for collaborative plan creation, co-tasking for guiding execution, action guards for sensitive operations, and plan learning/retrieval to improve future automation. It supports integration with Microsoft's Fara-7B model and offers flexible configuration for various LLM clients like Azure OpenAI and Ollama, making it a versatile platform for studying human-agent interaction.

Alrite is a company dedicated to developing innovative AI applications, with a current focus on tools such as Rizzpad and PetCoco. These applications are crafted to streamline and enhance various aspects of daily life for users. Alrite emphasizes creating user-friendly AI solutions that are accessible and beneficial for a wide range of consumer applications. The company aims to integrate artificial intelligence seamlessly into everyday routines, making advanced technology practical and easy to use for everyone.

Fuji-Web is an intelligent AI agent designed to automate web-based tasks directly from your browser's sidepanel. It understands user intent, navigates websites autonomously, and executes tasks on your behalf, providing explanations for each action taken. This transparency allows users to maintain control while leveraging AI for efficiency. The tool is installed as a browser extension, requiring an OpenAI or Anthropic API key for functionality. It supports complex and cross-tab workflows, with future plans for integration with browser automation frameworks like Puppeteer and Playwright, as well as features for saving and sharing workflows. Fuji-Web is open-source, allowing users to build the extension from source.

LookaheadDecoding is an open-source project designed to significantly accelerate Large Language Model (LLM) inference by breaking the traditional sequential dependency of token generation. This innovative approach utilizes a parallel decoding algorithm, eliminating the need for a draft model or a separate data store. Motivated by Jacobi decoding, LookaheadDecoding collects and caches n-grams from Jacobi iteration trajectories, enabling simultaneous processing of future tokens. The process is divided into a lookahead branch, which generates new n-grams within a defined window, and a verification branch, which validates promising candidates. This method has demonstrated substantial latency reductions, achieving speedups ranging from 1.5x to 2.3x on various datasets and models. The tool supports sampling and FlashAttention, and is implemented with an attention mask to maximize GPU parallel computing power, making it a valuable resource for optimizing LLM performance.

rag-in-action is a comprehensive open-source code repository and training program focused on end-to-end RAG (Retrieval-Augmented Generation) system design, evaluation, and optimization. It breaks down RAG into 10 core components, offering practical projects to master the entire RAG workflow. The resource emphasizes tailoring RAG solutions to specific business needs and scenarios, rather than a one-size-fits-all approach. It covers modules from data loading and text chunking to vector embedding, retrieval processing, indexing, response generation, and system evaluation. The project supports both LangChain and LlamaIndex frameworks, with detailed environment configurations for GPU and CPU versions across Ubuntu, MacOS, and Windows.

MCP-for-security provides a collection of Model Context Protocol (MCP) server implementations for various security testing tools, enabling their integration into AI workflows. This open-source project, developed by Cyprox, aims to combine artificial intelligence with security tools for advanced threat detection and automated responses. It supports a wide range of popular security tools including Amass, SQLmap, FFUF, NMAP, Masscan, and more, making them accessible through a standardized interface. Users can deploy these MCP servers via Docker or through manual setup, allowing for flexible integration into existing security and AI infrastructures. The project emphasizes community-driven development, speed, precision in automated threat detection, and a secure, transparent platform built on open standards.

Matterport3DSimulator is an AI research platform designed for deep reinforcement learning, computer vision, natural language processing, and robotics. It allows AI agents to interact with real 3D environments using visual information derived from panoramic RGB-D images. The simulator is based on the Matterport3D dataset, featuring 90 diverse indoor environments. Key capabilities include outputting real RGB and depth images, customizable image resolution and camera parameters, and support for off-screen rendering. It offers both C++ and Python APIs and is highly efficient, capable of around 1000 fps RGB-D off-screen rendering. The platform also includes the Room-to-Room (R2R) navigation dataset and task for training agents to follow natural language instructions.

Falcondale specializes in developing applied quantum machine learning and optimization solutions designed to deliver real-world impact. The company focuses on leveraging quantum intelligence to solve complex problems across various industries. Falcondale aims to provide a competitive edge through its advanced quantum technologies, offering solutions that go beyond traditional computational methods. Their expertise lies in translating cutting-edge quantum research into practical, deployable applications for businesses and organizations seeking innovative data analysis and optimization capabilities.

Autotab is a general AI agent designed to automate repetitive tasks end-to-end with superhuman reliability. It learns workflows by observing human demonstrations, similar to how one would teach a human teammate. Autotab operates within its own secure, local browser, allowing it to navigate complex applications, collect data, fill out forms, and take actions such as sending messages or triggering refunds. It can be deployed in Fortune 500 companies and tech-forward businesses to scale operations where hiring and onboarding are bottlenecks. Users can teach Autotab specific workflows via video messages or prepared documents, and it can run these tasks 24/7, on demand, on a schedule, or triggered via API.

Peekaboo is a powerful macOS command-line interface (CLI) tool and optional MCP server designed to empower AI agents with advanced screen capture and automation capabilities. It provides high-fidelity screen captures of applications or the entire system, including pixel-accurate captures with Retina 2x scaling. AI agents can leverage Peekaboo's natural-language interface to chain various tools like seeing, clicking, typing, scrolling, and hotkey presses, enabling comprehensive GUI automation. The tool supports multi-provider AI models such as OpenAI's GPT-5.1, Anthropic's Claude 4.x, xAI's Grok 4-fast, Google's Gemini 2.5, and local Ollama models for visual question answering. It's ideal for developers and technical users looking to create configurable, testable workflows with reproducible sessions on macOS.

AMA - Medical AI is an iOS application designed to offer personalized health and wellness information. It serves as a comprehensive AI assistant for health, fitness, nutrition, and diet-related inquiries, aiming to provide precise, fast, and personalized answers. Users can download the app from the App Store and leverage its AI capabilities to better understand their health and make informed decisions to improve their well-being. The tool focuses on delivering accurate and reliable health information, answering specific questions, and offering tailored health advice, making it a valuable resource for individuals seeking to enhance their health knowledge.

ImageToText.info is a free online OCR tool designed to accurately extract text from various image formats, including JPG, PNG, GIF, and PDF. Leveraging advanced AI technology, specifically tesseract-ocr, it offers high accuracy in converting visual text into editable digital formats. Users can upload, drag-and-drop, or paste image URLs to quickly convert single or batch images. The tool supports over 20 languages, allowing for diverse text extraction needs. Extracted text can be downloaded as a text file or copied to the clipboard, making it convenient for editing or integration into other documents. ImageToText.info emphasizes user privacy, stating no data is transmitted or stored, and offers a simple, registration-free experience for quick text extraction.

Deep-Learning-in-Production is a comprehensive GitHub repository curated by ahkarami, designed to serve as a valuable resource for deploying deep learning-based models in production environments. The repository compiles useful notes and references across various deep learning frameworks, including PyTorch, TensorFlow, Keras, and MXNet. It covers essential topics such as model conversion (e.g., PyTorch to C++, Keras to C++), model serving with tools like Flask, TorchServe, and TensorFlow Serving, and deployment on platforms like AWS Lambda and Kubernetes. Additionally, it provides insights into model quantization, speed optimization, and general deep learning deployment toolkits like OpenVINO and NVIDIA Triton Inference Server. The repository also includes resources for front-end and back-end development, mobile/embedded device deployment, and MLOps, making it a holistic guide for machine learning engineers and data scientists looking to operationalize their models.

This tool entry describes a robust security framework for managing multiple AI agents using a single markdown file per agent. It outlines how Ultrathink, an e-commerce store run autonomously by AI agents, governs its eight specialized agents. The core of the system involves defining per-role tool restrictions in YAML frontmatter within each agent's markdown instruction file, limiting what each agent can access, modify, or destroy. A shared CLAUDE.md file establishes project-wide rules that all agents inherit, ensuring hard constraints like mandatory security reviews. The system also incorporates daily automated audits performed by a security agent, which reviews instruction files and code changes to catch vulnerabilities and capability creep. This file-based governance prioritizes rapid evolution and auditability over cryptographic signing for internal systems.

The live website content for SapienAPI is entirely in Chinese and primarily displays information related to industrial equipment, such as various types of saws, cutting machines, and related accessories. There is no discernible information or mention of AI, search engines, or any related technology. The meta tags and homepage content are also in Chinese, focusing on industrial products and contact information for a company in Shijiazhuang. The original description of SapienAPI as an AI-powered search tool utilizing LLMs and real-time web data to find websites is not supported by the current live website content.

Acumino provides AI-powered robot models specifically designed for dexterous industrial automation. By training its AI on extensive robot interaction data, Acumino facilitates the seamless deployment of intelligent robot workers capable of performing complex tasks with high precision. This technology is engineered to offer scalable, reliable, and cost-efficient solutions, significantly enhancing operational efficiency and unlocking substantial return on investment in various industrial environments. Acumino's focus is on transforming manufacturing and logistics through advanced robotics.

Bonza.Chat is an advanced AI platform designed for creating and interacting with personalized virtual AI companions. Users can customize their AI's appearance, personality traits, communication style, and interests to craft their ideal digital partner. The platform supports uncensored conversations, remembers past interactions for a more natural experience, and offers features like AI image generation. It functions directly in a web browser, making it accessible across various devices without requiring app downloads. Bonza.Chat provides a free plan for basic chat and offers premium subscriptions for unlimited messaging, advanced features, and NSFW content, focusing on emotional connection and personalized digital relationships.

DeepTutor is an agent-native personalized learning assistant designed to enhance the educational experience through adaptive and intelligent tutoring. It features a unified chat workspace with six modes, including Deep Solve, Quiz Generation, Deep Research, Math Animator, and Visualize, all sharing the same context. The AI Co-Writer acts as a first-class collaborator in a multi-document Markdown workspace, drawing from your knowledge base and the web to rewrite, expand, or summarize text. Its Book Engine compiles structured, interactive "living books" with 14 block types, such as quizzes, flashcards, and interactive demos. DeepTutor also includes a Knowledge Hub for building RAG-ready knowledge bases from various document types and persistent memory that builds a living profile of the user's learning journey. Personal TutorBots offer autonomous tutoring with their own memory, personality, and skill sets, evolving with the user.

DeepSeek-V3 is a powerful Mixture-of-Experts (MoE) language model featuring 671B total parameters, with 37B activated for each token, ensuring efficient inference and cost-effective training. Building on the DeepSeek-V2 architecture, it introduces an innovative auxiliary-loss-free strategy for load balancing and a multi-token prediction training objective for enhanced performance. The model was pre-trained on 14.8 trillion diverse tokens and further refined through Supervised Fine-Tuning and Reinforcement Learning. DeepSeek-V3 demonstrates superior performance against other open-source models and rivals top closed-source alternatives, particularly excelling in math and code tasks. It supports local deployment on various hardware and open-source community software, including SGLang, LMDeploy, and TensorRT-LLM, with options for FP8 and BF16 inference.

StreamingLLM is an innovative open-source framework designed to address the challenges of deploying Large Language Models (LLMs) in streaming applications that require processing infinite-length inputs. It introduces the concept of "attention sinks" to efficiently manage Key and Value (KV) states, allowing LLMs to generalize to infinite sequence lengths without fine-tuning. This approach prevents the performance degradation seen in traditional window attention methods when text length exceeds cache size. StreamingLLM enables models like Llama-2, MPT, Falcon, and Pythia to perform stable and efficient language modeling with millions of tokens, offering up to a 22.2x speedup over sliding window recomputation baselines. It is particularly optimized for scenarios such as multi-round dialogues where continuous operation without extensive memory or dependency on past data is crucial.

micro_diffusion is an open-source repository from Sony Research that provides a minimalistic implementation for training large-scale diffusion models from scratch with an extremely low budget. Utilizing only 37 million publicly available real and synthetic images, it can train a 1.16 billion parameter sparse transformer for approximately $1,890, achieving a strong FID score on the COCO dataset. The repository includes training code, dataset code, and pre-trained model checkpoints for off-the-shelf generation. It supports progressive training from low to high resolution and incorporates patch masking for performance optimization and reduced training time.

open-llms is a comprehensive GitHub repository that serves as a curated list of open Large Language Models (LLMs) explicitly licensed for commercial use, including Apache 2.0, MIT, and OpenRAIL-M. This resource is invaluable for developers, researchers, and businesses looking to integrate open-source LLMs into their applications without licensing concerns. The repository details each model's release date, available checkpoints, associated research papers or blog posts, parameter sizes, context lengths, and specific licenses. It also includes a dedicated section for open LLMs tailored for code generation, offering insights into models like SantaCoder, CodeGen2, and StarCoder. Contributions to the list are welcomed, ensuring it remains up-to-date with the latest commercially viable open LLM releases.

StreamingVLM is an innovative AI tool designed for real-time understanding of effectively infinite video streams. Developed by mit-han-lab, it addresses common challenges in long-video analysis by maintaining a compact KV cache and aligning training directly with streaming inference. This approach efficiently avoids the quadratic cost associated with traditional methods and mitigates the pitfalls of sliding-window techniques. The system is capable of running at up to 8 frames per second (FPS) on a single H100 GPU, offering stable and efficient video processing. It has demonstrated superior performance, winning 66.18% against GPT-4o mini on a new long-video benchmark and also enhances general Video Question Answering (VQA) capabilities without requiring task-specific fine-tuning. The project provides scripts for environment setup, inference, supervised fine-tuning (SFT), and various evaluations including OVOBench and VQA tasks.