AI Agents & Automation

acu is a comprehensive, curated list of resources dedicated to AI agents for computer use. It serves as a central hub for researchers, developers, and enthusiasts looking to explore autonomous programs capable of reasoning, planning, and acting within digital environments. The repository categorizes resources into articles, research papers, surveys, frameworks, models, UI grounding techniques, datasets, benchmarks, and safety considerations. It covers a wide range of topics from foundational models and reinforcement learning approaches to specific applications in web navigation, mobile device control, and GUI automation. This resource is invaluable for understanding the current landscape and future directions of AI agents interacting with computer interfaces.

The Ghostery adblocker is an open-source JavaScript library designed for efficient blocking of ads, trackers, and other annoyances. It boasts strong compatibility, supporting 99% of filters from popular projects like Easylist and uBlock Origin. This library is highly embeddable, allowing developers to integrate adblocking functionality into various environments including Node.js, Puppeteer, Electron, and WebExtension (Chrome and Firefox). It is the core technology powering adblockers from Ghostery and Cliqz, used by millions of users, and has been battle-tested in diverse use cases such as mobile-friendly adblockers in React Native, web extensions, and batch request processing in Node.js. Its innovative algorithms are recognized for their performance, influencing other adblockers like Brave.

AgentField is an open-source control plane designed to build, run, and scale AI agents as production-ready APIs and microservices. It allows developers to write agent logic in Python, Go, or TypeScript, transforming it into robust infrastructure with features like routing, coordination, memory, async execution, and cryptographic audit trails. Every function becomes a REST endpoint, and each agent receives a cryptographic identity, ensuring every decision is traceable. AgentField supports structured AI output, human-in-the-loop approvals, cross-agent calls, and canary deployments. It also offers comprehensive observability, identity and access management, and verifiable credentials for tamper-proof execution receipts.

AgentDock is an open-source, backend-first framework designed for building sophisticated AI agents capable of executing complex tasks with configurable determinism. It comprises AgentDock Core, a framework-agnostic and provider-independent backend, and an Open Source Client, a Next.js application serving as a reference implementation. Built with TypeScript, AgentDock prioritizes simplicity, extensibility, and predictable AI system behavior. Key features include a modular node-based architecture, tools as specialized nodes, and comprehensive type safety. It supports both deterministic workflows and non-deterministic agent behaviors, allowing developers to balance creative AI capabilities with predictable system outcomes. AgentDock Pro, a cloud platform with visual workflow builders and advanced orchestration, is also coming soon.

Lirio is a personalization engine for digital health that leverages behavioral science and AI to power hyper-personalized health experiences. The platform helps organizations orchestrate each person’s unique healthcare journey through dynamic, personalized communications, which they call Precision Nudging®. By understanding and influencing human behavior, Lirio aims to remove barriers to health improvement and support individuals on their path to change. It moves beyond standard demographic segmentation to create highly tailored communications, fostering sustained engagement and healthier populations. The platform continuously learns and adapts behavioral interventions in real-time, helping to close gaps in care and reduce the total cost of care by automating ongoing touchpoints at scale.

Milo AI is a secure AI data analyst designed for business users to instantly analyze data using natural language. It connects to various databases, data warehouses, and tools like HubSpot and Salesforce, allowing users to ask questions as they would a colleague. Milo understands the query, finds answers across connected data sources, and presents them as charts, answers, and insights. The platform is built on a zero-trust architecture, ensuring data security and privacy with SOC 2, ISO 27001, and GDPR certifications. It supports over 700 integrations and offers enterprise-grade controls, regional data residency, and zero data retention, making it a reliable solution for real-time business intelligence without the need for analysts or complex BI tools.

async-rl offers a practical implementation of asynchronous 1-step Q learning, as detailed in the paper "Asynchronous Methods for Deep Reinforcement Learning." This open-source project leverages Tensorflow and Keras for deep Q network definition and optimization, while integrating with OpenAI's Gym library for interaction with the Atari Learning Environment. A key feature is its use of multiple actor-learner threads, which helps stabilize the learning process without relying on memory-intensive experience replay, making it efficient for machines with less RAM. The repository also includes a work-in-progress asynchronous advantage actor-critic implementation and provides instructions for training, visualizing with TensorBoard, and evaluating models.

Metaflow is an open-source, human-centric framework designed to assist scientists and engineers in building, managing, and deploying real-life AI and ML systems. Originally developed at Netflix and now supported by Outerbounds, it streamlines the entire development lifecycle, from rapid prototyping in notebooks to reliable, maintainable production deployments. Metaflow offers a simple, pythonic API covering foundational needs such as rapid local prototyping, experiment tracking, versioning, and visualization. It enables effortless scaling horizontally and vertically in the cloud, utilizing both CPUs and GPUs, with fast data access for massive compute workloads. Metaflow also simplifies dependency management and provides one-click deployment to highly available production orchestrators with built-in reactive orchestration. It powers thousands of AI and ML experiences across diverse companies, including Amazon, Doordash, and Netflix.

Mem0 offers a drop-in memory infrastructure designed for AI agents and applications, ensuring context persists across sessions and agents. It allows AI apps to continuously learn from past user interactions, leading to enhanced intelligence and personalization. Key features include a Memory Compression Engine that condenses chat history to reduce token costs and latency, and a system that learns and retrieves key memories as users interact. Mem0 is built for developers, offering SDK integrations for Python and NodeJS, and is backed by Y Combinator. It supports various use cases such as healthcare, education, e-commerce, customer support, and sales & CRM, providing adaptive memory solutions for diverse domains. The platform emphasizes enterprise-grade governance, reliability, and observability, with SOC 2 and HIPAA compliance.

Related AI functions as an intelligent conversational agent, designed to help users explore interconnected topics and generate contextually relevant responses. It leverages advanced AI to understand user queries, providing insights and content that expand upon initial ideas, thereby fostering deeper understanding and creative exploration. The tool offers both basic and premium AI models, web search functionality, reasoning capabilities, and the ability to upload files for contextual understanding. It is available with a free tier for basic use and a paid tier that unlocks full access to all features.

2022-Machine-Learning-Specialization is an open-source GitHub repository offering comprehensive materials for Andrew Ng's 2022 Machine Learning Specialization. This resource provides course code, test content, and slides across three main parts: Supervised Machine Learning (Regression and Classification), Advanced Learning Algorithms, and Unsupervised Learning (Recommenders, Reinforcement Learning). It encourages community contributions through pull requests for supplementing learning files and optimizing markdown notes, fostering a collaborative learning environment. The repository also includes environment configuration instructions for Python dependencies, making it accessible for students and developers to set up their learning environment.

awesome-multi-modal-reinforcement-learning is an open-source repository offering a curated and continuously updated list of resources for Multi-Modal Reinforcement Learning (MMRL). This collection primarily consists of research papers, including those from top conferences like NeurIPS, ICML, ICLR, CVPR, and CoRL, as well as pre-print archives like ArXiv. The repository aims to track the frontier of MMRL, focusing on agents that learn from diverse data types such as video (images) and language (text). While some included papers may not be strictly RL-focused, they are deemed useful for MMRL research. The resource is ideal for researchers and developers looking to stay informed on the latest advancements in multi-modal AI.

Chatronix is a multi-AI conversational platform designed to streamline the process of generating effective AI prompts. Users gain instant access to a comprehensive library of over 550 categorized, ready-to-use prompts covering diverse domains such as social media marketing (SMM), copywriting, education, business, and general marketing. This eliminates the need to spend time crafting prompts from scratch. Beyond the extensive library, Chatronix also includes an AI Prompt Generator, allowing users to create custom, high-quality prompts tailored to their specific requirements. The platform aims to enhance productivity and efficiency for individuals and businesses leveraging AI for content creation and strategic communication.

Chaperoned offers a secure and private platform designed for users to interact with AI agents. This tool emphasizes personalized chat experiences, giving individuals the ability to customize their interactions to align with specific preferences and needs. It aims to create a controlled environment where users can explore the capabilities of AI agents without concerns about data privacy or security. The platform focuses on delivering a tailored experience, ensuring that each user can adapt the AI's responses and behavior to suit their unique requirements, making it suitable for various applications where controlled and personalized AI interactions are crucial.

Plataine offers AI-based optimization solutions specifically designed for advanced manufacturing, providing intelligent, connected digital assistants for production floor management and staff. The platform empowers manufacturers to optimize decisions, meet production deadlines, and grow with AI superpowers. Key features include the Production Scheduler, which creates optimal schedules with one click, and the Material Asset Tracker for real-time tracking of assets. Plataine's solutions aim to increase throughput, improve on-time delivery, reduce material waste, and enhance overall operational efficiency. It supports various industries such as Aerospace & Defense, Automotive, Composite Material Manufacturing, and Industrial Manufacturing, helping businesses achieve higher levels of sustainability and maximize ROI through improved visibility and traceability.

DI-engine is a generalized decision intelligence engine built for PyTorch and JAX, offering a comprehensive framework for reinforcement learning. It features python-first and asynchronous-native task and middleware abstractions, integrating key decision-making concepts like Env, Policy, and Model. The framework supports a wide array of deep reinforcement learning algorithms, including DQN, PPO, SAC, and multi-agent, imitation, offline, and model-based RL. Beyond algorithms, DI-engine aims to standardize decision intelligence environments and applications, catering to academic research and prototype development. It also includes highly re-usable modules for RL optimization, PyTorch utilities, and system optimizations for efficient large-scale RL training.

DR Tulu is an open-source Deep Research (DR) model designed for tackling long-form research tasks. The DR Tulu-8B model has demonstrated performance comparable to OpenAI DR on long-form DR benchmarks. This repository provides the official code for DR Tulu, including an agent library with a MCP-based tool backend, high-concurrency async request management, and a flexible prompting interface for developing and training deep research agents. It also includes RL training code based on Open-Instruct and SFT training code based on LLaMA-Factory, allowing for supervised fine-tuning and reinforcement learning with GRPO and evolving rubrics. An interactive CLI demo is available for users to experiment with DR Tulu-8B.

drl_grasping is an open-source project focused on advancing robotic manipulation through deep reinforcement learning. It enables robots to acquire robust grasping policies for diverse objects using compact 3D observations in the form of octrees. The project emphasizes sim-to-real transfer, allowing policies trained in simulation to be evaluated on real robots with zero-shot transfer. It includes multiple RL environments for robotic manipulation, supporting continuous actions in Cartesian space and various observation variants like RGB images, depth maps, and octrees. The framework is compatible with Gym API and has been tested with end-to-end model-free actor-critic algorithms like TD3, SAC, and TQC, with a setup for model-based algorithms also provided.

GaussianEditor is an open-source AI tool presented at CVPR 2024, designed for swift and controllable 3D editing leveraging Gaussian Splatting technology. It provides capabilities for adding, deleting, and modifying objects within 3D scenes, with editing operations completing in just 2-7 minutes. The tool offers both a WebUI, powered by Viser, and a command-line interface for flexible use. It supports various editing functionalities, including generating edited 2D images as guidance using InstructPix2Pix for modifications, and employing ControlNet-Inpainting for adding new elements. GaussianEditor is built upon a foundation of other advanced repositories like Wonder3D and Threestudio, making it a powerful solution for 3D content creation and manipulation.

GPS-Gaussian is an innovative AI tool that offers a generalizable pixel-wise 3D Gaussian representation for real-time human novel view synthesis. Developed as a highlight of CVPR 2024, this tool allows for instant generation of novel views of unseen characters without requiring any fine-tuning or optimization. It provides a robust solution for computer vision and graphics researchers and developers working with 3D human models. The repository includes detailed instructions for installation, dataset preparation, and training, supporting both synthetic and real-world data. Users can pretrain a depth prediction model and then train the full model, with options for offline disparity determination to speed up subsequent training. Pretrained models are also available for quick evaluation and testing.

gpt-2-output-dataset is an Open Source project by OpenAI providing a comprehensive dataset of GPT-2 outputs. It includes 250,000 documents from the WebText test set, alongside 250,000 random samples and 250,000 samples generated with Top-K 40 truncation for each GPT-2 model (small-117M, medium-345M, large-762M, xl-1542M). This dataset is specifically designed for research in areas such as detection of AI-generated text, biases, and more. It also offers samples from a GPT-2 model finetuned to output Amazon reviews, encouraging research into finetuned model detection. The project provides detectability baselines and a script for easy download of the data.

Latent-NeRF is an official implementation for "Latent-NeRF for Shape-Guided Generation of 3D Shapes and Textures," providing a robust framework for creating 3D content. It leverages Latent Diffusion Models for efficient text-guided 3D object generation, operating directly in a compact latent space to avoid repeated encoding. The tool offers three primary models: a purely text-guided Latent-NeRF, a Latent-NeRF with soft shape guidance for more precise control over the generated shape using abstract geometries (Sketch-Shapes), and Latent-Paint for generating high-quality textures on explicit 3D meshes. This unique combination of text and shape guidance significantly enhances control over the generation process, making it suitable for various 3D content creation tasks. It also supports Textual Inversion tokens for conditioning object generation on specific styles or objects.

kernel-memory is an open-source research project from Microsoft, providing a memory solution for users, teams, and applications. This tool represents a full rewrite of an initial prototype, incorporating lessons learned and exploring new ideas in the space. It is actively evolving and may change without notice, serving primarily as a learning resource rather than production-ready software. The project emphasizes areas such as content quality, privacy, and collaboration, with future developments being built using AI and Amplifier concepts. Users should be aware that it is experimental software with no stability or compatibility guarantees.

DeepAnalyze is the first agentic LLM designed for autonomous data science, capable of handling a wide range of data-centric tasks without human intervention. It supports the entire data science pipeline, including data preparation, analysis, modeling, visualization, and report generation. The tool can conduct deep research on diverse data sources, such as structured (Databases, CSV, Excel), semi-structured (JSON, XML, YAML), and unstructured data (TXT, Markdown), producing analyst-grade research reports. DeepAnalyze is fully open-source, with its model, code, training data, and demo all publicly available, allowing users to deploy or extend their own data analysis assistant. It offers multiple interfaces, including WebUI, JupyterUI, and a command-line interface, and provides an OpenAI-style API for integration.