AI Agents & Automation

Biomni is a general-purpose biomedical AI agent designed to autonomously execute a wide range of research tasks across diverse biomedical subfields. It integrates cutting-edge large language model (LLM) reasoning with retrieval-augmented planning and code-based execution, enabling scientists to dramatically enhance research productivity and generate testable hypotheses. Biomni supports various LLM providers like Anthropic, OpenAI, Azure OpenAI, Gemini, and Groq, and can be configured via environment variables or a .env file. It features a data lake for biomedical information, a Gradio interface for interactive use, and configuration management for consistent settings. Additionally, Biomni can generate PDF reports of execution traces, supports Model Context Protocol (MCP) for external tool integration, and includes a Know-How Library of best practices. It also offers Biomni-R0, a specialized reasoning model for biology, and Biomni-Eval1, a comprehensive evaluation benchmark.

cagent, developed by Docker Engineering, is an AI Agent Builder and Runtime designed for creating, running, and sharing intelligent AI agents. It leverages a declarative YAML configuration, eliminating the need for extensive coding. The platform supports a multi-agent architecture, enabling teams of specialized agents to collaborate and delegate tasks automatically. With a rich tool ecosystem, including built-in tools and integration with any MCP server, cagent offers flexibility. It is also AI provider agnostic, supporting major models like OpenAI, Anthropic, Gemini, AWS Bedrock, and Mistral. Key features include advanced reasoning capabilities with built-in think, todo, and memory tools, as well as pluggable RAG for retrieval. Agents can be packaged and shared via any OCI registry, making deployment and collaboration seamless.

Chrome-GPT is an experimental AutoGPT agent designed to take control of an entire Chrome session on your desktop. Utilizing Langchain and Selenium, it allows for interactive scrolling, clicking, and text input on web pages, enabling the AutoGPT agent to navigate and manipulate web content. Key features include Google search capabilities, long-term and short-term memory management, and various Chrome actions such as describing webpages, interacting with elements, and switching tabs. It supports multiple agent types, including Zero-shot, BabyAGI, and Auto-GPT, with planned support for Chrome plugins. Users should be aware of its experimental nature, potential for incorrect actions, and current limitations like slow response times and occasional parsing issues.

SPAICE OS is an advanced operating system designed to bring reliable spatial-AI autonomy to aircraft and satellites, even in challenging environments where GNSS or communications may fail. It transforms any aircraft or satellite into a Spatial Agent capable of understanding and operating autonomously using only onboard cognitive sensors. The system focuses on three core technological pillars: Perception, which turns raw sensor data into situational awareness; Planning, for computing optimal trajectories in real-time onboard; and Control, for executing smooth, reliable, and collision-free maneuvers. SPAICE is ideal for applications such as Intelligence, Surveillance & Reconnaissance, Command & Control, Distributed Intelligence, Target Detection, Classification and Tracking, Self-Localization in GPS-Denied Environments, and Terrain Mapping.

ClawX is a desktop application designed to bridge the gap between powerful AI agents and everyday users by providing a graphical interface for OpenClaw AI agents. It eliminates the need for command-line interaction, offering a seamless desktop experience for AI orchestration. Key features include one-click installation, visual settings for configuration, automatic gateway lifecycle management, and a unified panel for multiple AI providers. ClawX supports intelligent chat interfaces with rich content rendering, multi-channel management for independent AI tasks, and cron-based automation for scheduling AI tasks. It also boasts an extensible skill system with pre-built skills and secure integration with various AI providers like OpenAI and Anthropic, storing credentials in the system's native keychain. The application supports Windows, macOS, and Linux, and offers adaptive theming and startup launch control.

Claude-API offers an unofficial Python API for interacting with Claude AI, providing developers with the ability to integrate Claude's capabilities into their own applications and workflows. This project facilitates tasks such as sending messages, managing conversations, and handling file attachments programmatically. It supports functionalities like listing all conversations, sending messages with or without attachments, deleting conversations, retrieving chat history, creating new chats, resetting all conversations, and renaming chats. The API is designed for ease of use within Python environments, requiring only the `requests` library and a Claude AI cookie for authentication. It's an open-source solution, making it accessible for developers looking to build custom AI-powered applications.

claude-code-sub-agents offers a comprehensive collection of 33 specialized AI subagents designed to extend Claude Code's capabilities across the entire software development lifecycle. Each subagent acts as an expert in a specific domain, automatically invoked based on context analysis or explicitly called when specialized expertise is needed. Key features include intelligent auto-delegation, domain-specific expertise in various technologies, multi-agent orchestration for complex workflows, and built-in quality assurance. The tool is optimized for performance and covers areas like frontend, backend, mobile development, infrastructure, quality assurance, data engineering, AI/ML, and security. It also includes an 'agent-organizer' for master orchestration of complex, multi-agent tasks.

Soaring Titan specializes in building and deploying production agentic AI systems for businesses. They work with portfolio companies, growth-stage businesses, and organizations backed by investors to integrate AI for operating transformation, not just additive improvements. Their approach involves auditing workflows, data architecture, and integrations to identify where AI can compound value. They embed with teams to ship agentic systems designed to deliver measurable operating value within 100 days, then scale these repeatable playbooks across departments or portfolio companies. With a background in FinTech and AI since 2020, they emphasize building alongside clients rather than just advising, focusing on tangible outcomes and measurable impact.

AI Hustler is a specialized platform designed to bridge the gap between businesses and high-caliber AI talent. It offers access to both human AI freelancers and advanced AI agents, enabling companies to find tailored solutions for their AI initiatives. The platform emphasizes flexibility, allowing businesses to scale their AI projects based on evolving needs through a gig economy model. AI Hustler prioritizes security and reliability, ensuring a safe environment for on-demand AI collaboration. It features various AI services and projects, from AI-driven robotics to client communication and data analysis, catering to a diverse range of business requirements. The platform aims to empower startups and established enterprises to innovate and stay competitive by leveraging AI.

DeepSeek-Math-V2 is an advanced AI model specifically designed for mathematical reasoning, focusing on self-verifiable theorem proving. It addresses the limitations of traditional reinforcement learning by emphasizing rigorous step-by-step derivation rather than solely relying on final answer accuracy. The model trains a proof generator using an accurate and faithful LLM-based verifier, incentivizing the generator to identify and resolve issues in its own proofs. DeepSeek-Math-V2 demonstrates strong capabilities, achieving gold-level scores on IMO 2025 and CMO 2024, and a near-perfect score on Putnam 2024. This approach aims to push the limits of deep reasoning and advance mathematical AI systems.

DesktopCommanderMCP serves as an MCP server for Claude, enabling advanced terminal control, file system search, and diff file editing. This tool empowers AI to interact with your computer's file system and execute commands, going beyond typical AI editors. It supports various operations like reading/writing files (text, Excel, PDF, DOCX), creating/listing directories, and performing recursive searches. Developers and technical users can leverage its capabilities for code editing, process management, and automating tasks, all while using host client subscriptions to avoid API token costs. It offers multiple installation methods, including npx, bash scripts, Smithery, manual configuration, and Docker, catering to different user preferences and environments.

docling is an open-source tool designed to prepare documents for generative AI applications, streamlining document processing and providing seamless integrations. It handles a wide array of document formats including PDF, DOCX, PPTX, XLSX, HTML, WAV, MP3, images, LaTeX, and plain text. A key feature is its advanced PDF understanding, encompassing page layout, reading order, table structure, code, formulas, and image classification. docling offers a unified `DoclingDocument` representation and various export options like Markdown and JSON. It supports local execution for sensitive data and integrates with popular AI frameworks such as LangChain, LlamaIndex, Crew AI, and Haystack, making it suitable for developers and AI researchers working on document-related projects.

Docta is an advanced open-source data-centric AI platform designed to detect and rectify issues within various data types, including tabular, text, and image data, as well as pre-trained model embeddings. It aims to improve model performance by ensuring data health through diagnosis, curation, and nutrition services. The tool is training-free, making it a premium-free option that operates on user data without additional prerequisites. Docta can identify label errors, as demonstrated with LLM alignment data (e.g., Anthropic's HH-RLHF dataset) and real-world human-annotated image data like CIFAR-N. It also excels at detecting rare patterns in datasets, which can be crucial for enhancing data quality and model robustness. The platform provides diagnosis reports and suggests corrections, such as improved ratings for LLM responses.

dots.ocr is a powerful vision-language model designed for universal accessibility, capable of recognizing virtually any human script and performing multilingual document layout parsing. It achieves state-of-the-art performance in standard multilingual document parsing among models of comparable size. A key differentiator is its ability to convert structured graphics, such as charts and diagrams, directly into SVG code, as well as parsing web screens and spotting scene text. The tool offers models like dots.mocr and dots.mocr-svg, with detailed evaluation benchmarks against other leading models. It provides flexible deployment options including vLLM inference for high performance and Hugging Face inference, making it suitable for developers and researchers working with complex document analysis tasks. The tool also supports parsing both image and PDF files, outputting structured JSON data, processed Markdown files, and layout visualizations.

DeepResearchAgent is an open-source, hierarchical multi-agent system designed for both deep research tasks and general-purpose problem-solving. The framework utilizes a top-level planning agent to orchestrate multiple specialized lower-level agents, enabling automated task decomposition and efficient execution across diverse and complex domains. Built on Autogenesis, a self-evolution protocol, it allows agents to dynamically instantiate, retrieve, and refine resources, improving during execution. Key components include agents for runtime logic, tools for callable capabilities, environments for stateful interfaces, memory systems for summarization, and optimizers for self-improvement. It emphasizes composability, inspectability through structured traces, and evolvability via explicit optimizers and persistent memory.

DiffIR is an efficient diffusion model specifically designed for various image restoration tasks, including super-resolution, inpainting, and deblurring. This project is the official implementation of the 'Diffir: Efficient diffusion model for image restoration' paper presented at ICCV2023. Unlike traditional diffusion models that are often inefficient for image restoration due to massive iterations, DiffIR employs a compact IR prior extraction network (CPEN) and a dynamic IR transformer (DIRformer) to achieve accurate estimations with fewer iterations. It offers pre-trained models and training/testing codes for different tasks, allowing users to improve image quality effectively and stably.

ECANet is an open-source implementation of the Efficient Channel Attention (ECA) module designed for Deep Convolutional Neural Networks (CNNs). This tool addresses the trade-off between performance and complexity in channel attention mechanisms by proposing a lightweight yet effective module. It avoids dimensionality reduction and uses an efficient 1D convolution for local cross-channel interaction, adaptively determining the kernel size. ECANet demonstrates clear performance gains with only a handful of parameters, making it highly efficient. It has been extensively evaluated on image classification, object detection, and instance segmentation tasks, showing favorable results against existing counterparts while maintaining low computational overhead.

eda_nlp is an open-source tool designed for data augmentation in Natural Language Processing (NLP), specifically aimed at improving performance on text classification tasks. Presented at EMNLP 2019, it offers a generalized set of easy-to-implement techniques that have shown substantial improvements, particularly on datasets with fewer than 500 samples. Unlike methods requiring extensive language model training, eda_nlp focuses on simple text editing operations. Key techniques include Synonym Replacement (SR), Random Insertion (RI), Random Swap (RS), and Random Deletion (RD). The tool is straightforward to use, requiring NLTK installation and a simple command-line interface to augment text data in a label-sentence format.

AyGLOO specializes in applying artificial intelligence to solve real-world business problems, creating tailored solutions that combine automation, language comprehension, and ethical responsibility. Their services include designing and implementing Agentic AI systems for autonomous task automation and information analysis, as well as Prescriptive Decision AI, which evaluates prediction reliability and calculates the expected impact of actions. AyGLOO's approach ensures that AI systems are explainable, traceable, and auditable, providing tangible results for clients across various sectors. They have a proven track record with projects for companies like Bidafarma, Suzuki, and PwC, demonstrating their ability to transform businesses through AI.

Gemini-OpenAI-Proxy acts as a crucial bridge, enabling applications designed for the OpenAI API to interact directly with Google's Gemini Pro protocol. This proxy facilitates seamless communication for key functionalities including Chat Completion, Embeddings, and Model endpoints. It offers straightforward deployment via Docker and allows users to integrate their Google AI Studio API key as if it were an OpenAI key. The tool also provides model mapping for various GPT models to their Gemini counterparts, with an option to disable mapping for direct Gemini model access. While Google AI Studio now offers an official OpenAI-compatible API endpoint, this proxy remains a viable solution for specific integration needs.

Frigate is a comprehensive, local NVR solution specifically designed for integration with Home Assistant, featuring advanced AI object detection capabilities. It leverages OpenCV and TensorFlow to perform real-time object detection directly on local IP camera feeds. The system is engineered for minimal resource consumption and maximum performance, employing low-overhead motion detection to trigger object detection only when necessary. Frigate utilizes multiprocessing to ensure real-time processing and communicates via MQTT for seamless integration with other systems. It supports 24/7 recording with retention settings based on detected objects, re-streaming via RTSP, and offers WebRTC & MSE for low-latency live viewing. Use of a GPU or AI accelerator is strongly advised for optimal performance.

The fraud-detection-handbook is a comprehensive, open-source resource dedicated to reproducible machine learning for credit card fraud detection. It functions as a practical handbook, offering detailed insights into the motivations and active research within this field. The resource emphasizes reproducibility, with all techniques and results provided in Jupyter notebooks that can be executed locally or on cloud platforms like Google Colab or Binder. It is designed for students and professionals interested in credit card fraud detection from a practical standpoint, as well as data practitioners and scientists dealing with sequential data and imbalanced classification problems. The handbook covers topics such as book overview, background, getting started, performance metrics, model selection, imbalanced learning, and deep learning.

GPT-4V-Act is an AI agent that leverages GPT-4V(ision) and a web browser to interact with web user interfaces, mirroring human operations through screen feedback and low-level mouse/keyboard interaction. Its primary objective is to facilitate a smooth transition between human and computer operations, enhancing UI accessibility, automating workflows, and enabling automated UI testing. The tool utilizes Set-of-Mark Prompting and a tailored auto-labeler that assigns unique numerical IDs to interactable UI elements. This allows GPT-4V-Act to deduce subsequent actions based on a task and a screenshot, using numerical labels for precise pixel coordinates for mouse/keyboard output. The project also incorporates features like JS DOM auto-labeler, clicking, and typing characters.

TensorFlow Fold is a specialized library designed for creating TensorFlow models that can process structured data with dynamic computation graphs. This means the structure of the computational graph can change based on the input data, making it highly adaptable for tasks like sentiment analysis on parse trees of varying shapes and sizes. A core feature is dynamic batching, which transforms batches of arbitrarily shaped computation graphs into a static graph. This static graph maintains consistent structure regardless of input, allowing for efficient execution within TensorFlow. The library leverages low-level APIs like Loom to create TensorFlow operations such as concat, while_loop, and gather, optimizing performance. It is particularly useful for researchers and developers working with complex, variable-structure data in deep learning applications.