All published AI news and analysis.

NVIDIA has announced Nemotron 3.5 Content Safety, a customizable multimodal AI safety solution designed for enterprise applications. It aims to help businesses manage and mitigate risks associated with generative AI, offering features for detecting harmful content across various modalities.

Hugging Face has updated its command-line interface (CLI) to better support AI agents interacting with the Hugging Face Hub. The redesign focuses on making the CLI more agent-friendly, enabling automated workflows for tasks like model management and dataset access.

Castor is a new self-hosted AI agent designed for business workflows, offering flexibility in LLM integration and data privacy. It supports various OpenAI-compatible LLMs and can be deployed on local hardware or cloud infrastructure.

SlideRule is an open-source product rehearsal engine that transforms a single sentence into a runnable product plan. It emphasizes transparency and deterministic gates, allowing developers to visualize and execute each step of product development.

HotPlex is a self-hosted gateway that unifies access to various AI coding agents, enabling them to be used across platforms like Web, Slack, and Feishu. It provides a consistent interface and manages sessions, persistence, and observability for team-wide AI agent integration.

NVIDIA has provided guidance on fine-tuning its Nemotron 3.5 ASR model, enabling developers to adapt the automatic speech recognition system for specific languages, domains, or accents. This process leverages Hugging Face tools and NVIDIA's NeMo framework.

A new dataset, MarcoFurrer/swiss-building-law-rag-bench, has been released on Hugging Face, featuring evaluation benchmarks for Retrieval-Augmented Generation (RAG) systems. The dataset focuses on Swiss cantonal building law documents and includes multilingual Q&A pairs.

agent-lsp is a new open-source project that functions as a Model Context Protocol (MCP) server, designed to provide AI coding agents with enhanced code intelligence. It orchestrates existing Language Server Protocol (LSP) servers to enable features like type-aware navigation, blast-radius analysis, and pre-verified edits across multiple programming languages.

The HeyClaude project is a community-driven, curated registry for Claude AI workflow assets, including agents, MCP servers, skills, and tools. It aims to provide a human-reviewed, machine-readable catalog for developers building with Claude.

HarnessClaw Engine is a new LLM programming assistant built in Go, featuring a 5-phase query loop for processing requests. It supports WebSocket for UI-first streaming, multi-turn dialogues, tool calling, and skill extension.