⭐ 27,696 | Astryx is an open source design system originally built inside Meta, powering over 13,000 apps. It provides 150+ accessible components that can be used seamlessly by both humans and AI assistants using the exact same tooling and references.
This update officially stops supporting Python 3.10, requiring users to upgrade to Python 3.11 or newer. It also introduces the new OpenAIClassifier feature to help developers work with the Decisions API more easily.
Hugging Face Transformers version 5.23.0 introduces a variety of new multi-modal and language models, including Inkling2, Liquid AI's LFM3, and InternLM3. It also overhauls the model exporters, allowing users to seamlessly export and run models on optimized runtimes like ONNX Runtime and OpenVINO with significant speed improvements.
llama.cpp and ggml have released their v1.0.0 versions, marking an important milestone for running artificial intelligence locally across various hardware devices. The project team announces that future updates will strictly follow semantic versioning to protect downstream developers who rely on their libraries for private and cross-platform applications.
According to AWS Machine Learning: Amazon announces the public availability of SGLang-Neuron, enabling developers to serve frontier AI models and generate reinforcement learning rollouts on AWS Trainium chips. Developed by RadixArk in collaboration with Annapurna Labs, the integration lets users keep their familiar SGLang workflow while running natively on PyTorch.
Vidit-Ostwal released version 1.15.41, which introduces support for approved private URL hosts. This update resolves project dependency issues with PyMuPDF, improves the human-feedback resume workflow by retaining earlier step outputs, and refines the LiteLLM path.
According to AWS Machine Learning: Java teams can now build autonomous AI agents using Spring AI and Amazon Bedrock, combining familiar enterprise patterns with production-ready managed infrastructure.
⭐ 193,971 | This repository provides Claude's official skill implementations, which are folders of instructions, scripts, and resources that help Claude handle specialized workflows. It lets developers build reusable custom capabilities, automate complex tasks, and reference production-grade document processing code.
⭐ 11,830 | Auto Ontology is an open library by NVIDIA that builds a trusted context layer for enterprise data. It solves the problem of AI agents hallucinating or giving wrong answers by mapping out where data lives and what it means. It features automatic schema ingestion and explainable AI workflows for reliable text-to-SQL querying.
This update brings several key improvements, notably removing chromadb from the embedding function layer and adding support for jsonl in rag data discovery. It also fixes async hook issues and improves agent stability, making AI development more reliable and smooth.
This is a bug fix update for the AWS Neuron SDK, which is the official software development kit used by developers to build and run AI models on Amazon's custom machine learning chips. It mainly cleans up issues found in the previous version to keep your AI workloads running smoothly without unexpected hiccups.
This is a software version release for an open-source project. It brings minor updates, bug fixes, or new features as part of its ongoing development cycle.
This project is a tool that helps developers write safer and more reliable AI applications using LangChain by adding type safety. It solves the common headache of silent runtime bugs caused by data type mismatches in AI workflows. Key updates include dropping old Python versions to boost performance, and introducing a new classification decision model.
⭐ 21,545 | agent-desktop lets an AI agent see and operate desktop apps on macOS, Windows, and Linux through the OS accessibility tree. It avoids screenshots or pixel matching by returning JSON-based refs for UI elements, allowing agents to click, type, and verify actions reliably.
This update brings exciting performance boosts and fixes for running AI models locally. It introduces concurrent generation support for Apple's MLX framework and leverages memory mapping for Vulkan integrated GPUs, making your local AI run smoother and faster. Plus, it improves the ChatGPT CLI experience with better restart handling.
This pre-release update for tt-metal brings major performance improvements and bug fixes across its low-level kernels, core infrastructure, and neural network library. It significantly enhances hardware optimization for advanced AI models like Kimi-K3, Ideogram 4, and MiniMax-H3, making AI computation faster and more reliable on specialized hardware.
This release brings exciting updates to the AI model hub, including fresh additions like AG-Pose and LFM3. It also introduces important export changes where old python commands are replaced by cleaner CLI tools, alongside better quantization support for models like minilm_v2 to help things run smoother and faster.
This patch release brings important security updates and fixes for distributed training issues. It hardens the template sandboxing and ensures device IDs are correctly configured specifically for NCCL, helping developers run their AI models more securely and stably.
According to Clearpath Robotics: At ROSCon 2026, engineers introduced Proton, a lightweight and transport-agnostic communication protocol built for resource-constrained embedded robotics. Instead of forcing heavy robotics middleware onto simple microcontrollers, Proton provides efficient data exchange and seamless ROS 2 integration without complex networking overhead.
