The best new AI tools and the news behind them, every morning.
59 itemsclear
Models and libraries you can run yourself.
Penguin Mail is an open-source email client written in Rust, designed for Linux systems and featuring AI capabilities. The project provides users with an alternative email application built on modern technology with integrated artificial intelligence features.
Musubi released PolicyLM-1.7B, a lightweight decision model designed for real-time content moderation, available with open weights. The model represents a potential shift in how platforms can implement moderation at scale.
OpenTPU is an open-source AI accelerator project available on GitHub, developed by AI researchers to provide an alternative to proprietary tensor processing hardware.
Google releases EmbeddingGemma 2, an open-source multimodal embedding model designed to be lightweight and accessible. The model enables developers to create text and image embeddings for search, clustering, and retrieval tasks.
Reflection has released Beam, an open-weight AI model designed to compete with Chinese models while requiring less computational resources. The company targets enterprises and sovereign nations, positioning Beam as part of an "AI factories" product that allows institutions to train customized local AI systems using their own data.
Reflection releases Beam, a new open-weight AI model available for researchers and developers to download and deploy independently. The model represents the company's contribution to open-source AI development.
An open-source tool enables macOS users to remove Apple Intelligence data from their devices, addressing privacy concerns by allowing deletion of the approximately 12GB of cached AI processing files generated during system use.
An open-source macOS application enables semantic search across all photos and video frames using AI, allowing users to find content by describing what they're looking for rather than using keywords or metadata.
Aleph Alpha releases Kolibri, an open-weight large language model supporting German and English languages. The model is made available for community use and deployment.
Aleph Alpha has released Kolibri, an open-weight language model designed with sovereignty principles. The model is available for independent deployment and use.
Meta is open-sourcing Muse, a generative AI model, to encourage developers to integrate it into consumer devices like televisions and household appliances. The move aims to expand Muse adoption across various hardware categories beyond Meta's own products.
A developer created an open-source tool that generates LEGO building instructions using AI models like ChatGPT and Claude. The system works by having AI write code in LDraw, a low-level assembly language that describes LEGO brick placement and assembly.
Clef is an open-source platform for building decision models with a new reinforcement learning fine-tuning capability. The project enables developers to create and customize AI models for decision-making tasks using open-source tools.
Jeeves is an open-source reasoning system that enhances decision-making models through improved inference capabilities. The project appears to be built for analyzing and improving AI model decision patterns.
Nvidia has released an open-source tool designed to enhance security in AI systems. The tool aims to address vulnerabilities and improve the robustness of AI deployments across various applications.
Reladraw is a diagram language that offers a middle ground between auto-layout tools like Mermaid and manual drawing applications, giving users control over element placement while maintaining the efficiency of text-based definition.
Ollaya is an open-source project that builds on Ollama's framework to support Jev-style decision models. It enables running these specialized models locally without proprietary infrastructure.
Whiteboard is an open-source desktop IDE designed for collaborative software architecture, enabling humans and AI agents to work together in a shared workspace. Built by a YC W26 team, it addresses the need for thoughtful design workflows in software development.
Bend is a programming language designed to prevent AI-related errors through formal proof verification, with support for both CPU and GPU execution. The language aims to reduce mistakes in AI systems by enforcing correctness guarantees at the language level.
Researchers present a method for generating and adapting language model weights dynamically from live data, enabling models to operate with effectively infinite parameters by computing weights on-demand rather than storing them statically.
OpenSpec is a framework for defining and managing AI specifications with a focus on lightweight implementation and customizable configuration. It provides tools for structuring AI system requirements and parameters in a standardized format.
Researchers have developed a method to compress large language models to 1.58 bits per parameter, advancing ternary quantization techniques that reduce model size while maintaining performance. This breakthrough pushes the boundaries of extremely low-bit model compression for large-scale AI systems.
A 4 billion parameter AI model is trained to generate SQL query plans that execute 81% faster than PostgreSQL's default planner on benchmarked queries. The approach uses machine learning to optimize query execution strategies without modifying the database engine itself.
Salesforce and Nvidia released Koa, an open-weight reasoning model built on Nvidia's Nemotron, designed for sales, marketing, and customer-support automation. The model represents a shift toward specialized AI systems that can be deployed and customized outside centralized AI labs.
IBM and NASA have jointly launched an AI model designed to analyze lunar imagery and identify ice deposits and craters on the Moon's surface. The tool aims to support lunar exploration and resource mapping efforts.
Researchers have developed an AI tool that maps small molecules, revealing previously hidden patterns in molecular space. The tool advances understanding of chemical compounds and their properties at scale.
Ripple has introduced an AI Starter Kit for the XRP Ledger designed to enable developers to build AI agents that interact with blockchain payments. The kit provides resources for creating autonomous systems that can execute payment-related tasks on the XRP Ledger network.
NVIDIA has released an open-source collection of agent tools and skills designed for physical AI applications. The release provides developers with resources to build and deploy AI systems that interact with physical environments.
NVIDIA has released Ising, open AI models designed to accelerate development toward practical quantum computers. The initiative combines NVIDIA's expertise in AI and quantum computing to provide accessible tools for quantum research and application development.
