Brief

Google AI has introduced Vibe Coding XR, a new workflow designed to simplify the creation of interactive XR experiences. This system leverages Gemini's capabilities with the open-source XR Blocks framework to translate natural language prompts into functional, physics-aware WebXR applications for Android XR devices. The goal is to accelerate prototyping by allowing creators to quickly test intelligent spatial experiences without extensive coding knowledge, with applications deployable in under 60 seconds. Google plans to demonstrate Vibe Coding XR at ACM CHI 2026. AI

OpenAI has launched "Instant Checkout" within ChatGPT, enabling users to purchase products directly from merchants without leaving the chat interface. This feature is powered by the newly released Agentic Commerce Protocol, an open standard co-developed with Stripe. Initially available for U.S. users to buy from Etsy and soon Shopify merchants, the protocol aims to facilitate seamless AI-driven commerce by allowing AI agents, people, and businesses to collaborate on purchases. OpenAI is open-sourcing the protocol to encourage broader adoption and integration by developers and merchants. AI

AI startup Webhound has launched a research agent designed to automate the creation of web-scraped datasets based on natural language prompts. The agent, initially built on Claude 4 Sonnet, was re-engineered using Gemini 2.5 Flash and a multi-agent system to significantly reduce costs and improve reliability. This new architecture includes specialized agents for planning, searching, critiquing, and validating data, along with a text-based browser for efficient extraction. AI

A developer has created an open-source AI-powered web service that integrates FastAPI for APIs, Pydantic-AI for agent construction, and Model Context Protocol (MCP) servers for tools. The service allows users to query information from sources like Hacker News and web search, presenting ranked trend cards with summaries. It supports various local LLM configurations and is containerized with Docker for production deployment. AI

Hugging Face has released a guide detailing techniques to optimize the performance of large language models using the Transformers library. The blog post, inspired by OpenAI's open-source contributions, focuses on practical methods for accelerating inference and training. It covers strategies such as quantization, efficient attention mechanisms, and optimized kernels to help developers achieve faster results with their models. AI

Recall.ai has launched a new Desktop Recording SDK designed to simplify the integration of meeting recording capabilities into other applications. This SDK addresses the complexities of capturing high-quality audio and video, including speaker identification and clean video compositing, without requiring a bot to be present in the meeting. The company aims to provide developers with a robust infrastructure solution, drawing on their experience powering recording features for over 2000 companies and overcoming significant technical challenges in reliability and efficiency. AI

Together AI has appointed Mahadev Konar as its new SVP of Infrastructure Engineering to bolster its GPU cloud services. Konar, a key figure in Apache Hadoop's development and formerly VP of Infrastructure at Instacart, will lead efforts to enhance the reliability, performance, and scalability of Together AI's platform. The company aims to provide AI-native startups with a robust infrastructure, enabling them to focus on product development rather than managing complex GPU environments. AI

SandboxAQ has introduced SAIR, a new AI platform designed to accelerate pharmaceutical research and development. SAIR leverages AI-powered structural intelligence to analyze complex biological data, aiming to speed up the discovery of new drugs and therapies. The platform is expected to enhance the efficiency of R&D processes within the pharmaceutical industry. AI

Hugging Face has introduced ahead-of-time (AOT) compilation for its ZeroGPU Spaces, enabling faster inference speeds. This optimization technique compiles models before deployment, reducing latency and improving the overall user experience for those running models without dedicated GPUs. The feature aims to make AI model deployment more accessible and efficient on their platform. AI

Smooth has launched a new serverless browser agent API designed for reliability, speed, and cost-efficiency, claiming to be 7x cheaper and 5x faster than existing solutions. The API aims to simplify web automation tasks for developers by handling complexities like instant browser spin-up and CAPTCHA solving. Separately, ContextFort has introduced a tool to provide visibility and control over AI coding agents like Cursor and Claude Code, addressing security concerns about agents accessing sensitive files and credentials on developer machines. AI

April, a new voice-controlled AI assistant, has launched on the App Store to manage emails and calendars. The application allows users to dictate replies, summarize messages, and reschedule meetings hands-free. It utilizes Deepgram for speech-to-text and Eleven Labs for text-to-speech, with custom servers for Google integration. The developers are focusing on low latency and natural interaction, while also considering user feedback on safety features like a 'safe mode' for non-destructive operations. AI

Skope, a new billing system, has launched to support outcome-based pricing for software products, particularly targeting the burgeoning AI market. The platform allows companies to charge customers only when their software delivers a specific result, aligning incentives and reducing buyer risk. Skope aims to simplify the implementation of this pay-per-performance model, which was previously challenging to manage at scale. AI

Channel3, a startup founded by George and Alex, has launched an API designed to provide developers with a comprehensive database of internet products. The service addresses the difficulty of accessing clean, structured product data from various retailers, which is often protected by bot detection. Channel3 uses computer vision and LLMs to identify, normalize, and de-duplicate product listings across multiple vendors, offering a unified API for developers to integrate product recommendations and affiliate monetization into their applications. The platform supports text and image-based searches, provides product details like price and specifications, and aims to facilitate developer earnings through commissions. AI

Cyberdesk, a startup founded by Mahmoud and Alan, has launched a new tool designed to automate repetitive tasks within legacy Windows desktop applications. Their approach uses a deterministic computer use agent that learns workflows from natural language instructions, offering a more reliable alternative to traditional Robotic Process Automation (RPA) scripts. The agent can self-correct based on screen state and only resorts to expensive AI models when unexpected anomalies occur, making it both robust and cost-effective for industries like healthcare and accounting. AI

Replit has introduced App Storage, a new object storage solution designed to simplify the hosting and saving of large files like images, videos, and documents within applications. This feature integrates seamlessly with Replit's Agent capabilities, allowing users to build apps that handle diverse file types with built-in authentication and database connections for permission management. App Storage is intended for a wide range of applications, from client portals and recipe apps to document management systems and online course platforms, offering SDKs for both JavaScript and Python. AI

The mcp-use framework has been released, enabling developers to build applications that can connect to various large language models like ChatGPT and Claude. This framework allows for the creation of MCP Servers and MCP Apps, with SDKs available in TypeScript and Python. It also includes an MCP Inspector for testing and debugging, and a cloud deployment option for production environments. AI

A new infrastructure project called PHP-ORT aims to bring machine learning inference capabilities directly to PHP, the server-side language used by a significant portion of the web. This development seeks to empower millions of PHP developers to integrate AI features into their applications without relying on external services or switching programming languages. PHP-ORT provides a core Tensor API, a high-performance math library, and integrates with ONNX for direct inference, promising significant speedups. AI

Together AI has launched Together Evaluations, a new platform designed to help developers benchmark large language models for specific tasks. The service allows users to define custom benchmarks and utilize leading open-source LLMs as judges to assess model response quality. This approach aims to provide a faster and more flexible alternative to manual labeling or rigid automated metrics, with an early preview now available. AI

Model ML, a company co-founded by Chaz Englander, is developing AI infrastructure tailored for the financial services industry. Their platform utilizes purpose-built agents and applications to automate complex workflows, significantly reducing the time required for tasks like quarterly earnings summaries. This automation allows financial professionals to shift their focus from routine work to higher-value, judgment-based roles, prompting a re-evaluation of organizational structures to become AI-native. AI

ZeroEntropy has developed specialized AI models, including rerankers and embeddings, designed for production systems that prioritize speed and accuracy over generalist models. Their offerings, such as zembed-1 and zerank-2, aim to provide lower latency and higher accuracy for applications like Retrieval Augmented Generation (RAG). These models are available for integration into existing stacks and can be deployed on cloud platforms like AWS and Azure, with a focus on security and compliance standards. AI

Hugging Face is transitioning its model and dataset hosting platform, the Hugging Face Hub, away from Git Large File Storage (LFS) to Xet, a new version control system designed for large files. This move aims to improve performance and scalability for managing the vast amounts of data associated with AI models. The migration process is expected to be gradual, with users being notified and guided through the transition. AI

Cactus has released an open-source AI engine designed for mobile devices and wearables, prioritizing low latency and reduced RAM usage. The engine supports multimodal capabilities, including speech, vision, and language models, with an option to fall back to cloud-based models. It features NPU acceleration for energy efficiency and offers OpenAI-compatible APIs for integration into various applications. AI

BrowserOS has launched as an open-source browser designed for the AI era, integrating AI agents that can automate web tasks through natural language commands. It prioritizes user privacy and offers extensive customization by supporting over 11 AI providers, including popular options like Anthropic Claude, Google Gemini, and OpenAI, as well as local models. The browser is built on a Chromium fork, ensuring compatibility with existing Chrome extensions and offering a user-friendly experience for both general users and developers. AI

Hugging Face has detailed its infrastructure alerting system, emphasizing its role in maintaining production stability. The system is designed to provide timely notifications for critical issues, enabling rapid response and minimizing downtime. This approach ensures the reliability of their platform, which hosts a vast number of AI models and datasets. AI

Octelium has released a new open-source, self-hosted platform designed for secure access and deployment. It functions as a unified zero-trust solution, offering capabilities such as a remote access VPN, ZTNA, an alternative to ngrok and Cloudflare Tunnel, an API gateway, and an AI gateway. The platform supports identity-based access control and can be used for deploying containerized applications and managing homelab infrastructure. AI

Hugging Face has integrated its Transformers library with SGLang, an open-source language model serving system. This integration allows developers to leverage Hugging Face's extensive model hub directly within SGLang for more efficient model deployment and inference. The collaboration aims to simplify the process of serving large language models, making advanced AI capabilities more accessible to a wider range of users and applications. AI

Glowstick is a new Rust crate designed to enhance tensor manipulation by integrating shape checking directly into the type system. This approach aims to make tensor operations safer and more intuitive, particularly for developers working with machine learning frameworks. The project, currently in its pre-1.0 phase, offers features like dynamic dimension support and improved error messages, with plans to align with ONNX operations. AI

Together AI has launched two new products, Together Code Sandbox and Together Code Interpreter, aimed at improving the execution of AI-generated code. Together Code Sandbox offers customizable virtual machine environments for building development tools and agentic workflows, featuring rapid VM startup and scaling capabilities. Together Code Interpreter provides a simpler API for session-based Python code execution within these secure sandboxes, designed for straightforward use cases. AI

Together AI has launched Together Code Interpreter (TCI), an API designed to securely execute code generated by large language models. This tool addresses the limitation of LLMs being unable to run the code they produce, enabling developers to integrate and test code within agentic workflows. TCI creates sandboxed environments for code execution, returning results that can be fed back to LLMs for iterative improvement and richer user responses. The interpreter has also shown promise in accelerating reinforcement learning operations by automating code evaluation and unit testing during model training. AI

Tinfoil, a startup founded by researchers from MIT and Cloudflare, has launched a new service designed to provide verifiable privacy for AI workloads hosted in the cloud. The platform utilizes secure enclave technology, particularly NVIDIA's confidential computing capabilities on GPUs, to ensure that neither Tinfoil nor the cloud provider can access sensitive data processed by AI models. This approach aims to enhance AI privacy by replacing trust with provable security, enabling more complex AI applications that require private data. AI

HelixDB has launched as an open-source platform designed to consolidate multiple database types for AI applications. It aims to eliminate the need for separate databases for application logic, relational data, vectors, and graphs by offering a unified graph and vector data model that also supports KV, document, and relational formats. The platform includes a CLI for local instance management and a "helix chef" tool that can bootstrap projects and even build applications from a single description with the help of coding agents. AI

ParaQuery, a new startup, has launched a GPU-accelerated Spark and SQL data processing solution. The platform aims to offer cost and performance benefits over existing solutions like Google BigQuery. ParaQuery leverages NVIDIA's RAPIDS technology to enhance traditional data processing tasks, which the founder notes are often mistakenly believed to be limited to AI and graphics. AI

Replit has launched Replit Auth, a new service designed to simplify the integration of user login and management into applications. This feature allows developers to add secure authentication, including social sign-in options, with minimal effort by simply including it in their Replit Agent prompts. Replit Auth leverages existing infrastructure for enterprise-grade security and provides tools for managing user data and accounts directly within the Replit Workspace. AI

Exa has launched Websets, a new search engine that uses embeddings and agentic workflows to provide precise results from the web, presented in a database-like table format. The service aims to combat the decline in search quality by performing extensive embedding searches and then using LLMs to verify each result against complex queries. While the process can take significant time, Exa believes the accuracy and detailed verification are worth the wait, offering an alternative to traditional keyword-based search. AI

Arcee AI has migrated its specialized small language models (SLMs) from AWS to Together Dedicated Endpoints, seeking improved cost, performance, and operational agility. The company focuses on training efficient models under 72 billion parameters for specific tasks like coding and general text generation. Arcee AI also developed Arcee Conductor, an inference routing system that directs queries to the most suitable model, including third-party options like GPT-4.1 and Claude 3.7 Sonnet, to optimize cost and performance. AI

Raven is a new ecosystem of OCaml libraries designed for numerical computing, machine learning, and data science. It aims to provide type-safe alternatives to popular Python libraries such as NumPy, JAX, and PyTorch. The project includes modules for n-dimensional arrays, automatic differentiation, tokenization, neural networks, dataframes, and plotting, with the goal of building a robust scientific computing environment. AI

Hugging Face has released significant updates to its Gradio MCP (Multi-Client Proxy) servers, enhancing their capabilities for LLM deployment. These improvements focus on boosting performance and user experience, allowing developers to more effectively upskill their large language models. The updates include new features and optimizations designed to streamline the process of building and managing MCP servers for LLM applications. AI

Morphik has launched an open-source Retrieval-Augmented Generation (RAG) system designed for developers to integrate complex context into AI applications. The system aims to simplify the process by offering a unified solution for storing, representing, and searching unstructured and multimodal data, addressing the limitations of traditional RAG pipelines that struggle with visually rich documents. Morphik provides features like multimodal search, fast metadata extraction, and integrations with tools such as Google Suite and Slack, with a free tier available for users. AI

A new open-source project offers sandboxed Chrome browsers that can be run as Docker containers or on Unikraft unikernels. This setup is designed for browser automation, web agents, and testing AI agents that interact with the web. The unikernel implementation provides features like automated standby mode with state snapshotting and extremely fast cold restarts, enabling low-latency event handling. AI

Gradio is a Python library designed to simplify the creation of user interfaces for machine learning models. It allows developers to quickly build interactive demos and share them with others. The library offers features like pre-built UI components, easy integration with popular ML frameworks, and the ability to deploy applications with a single command. AI

AI startup mrge has launched a new platform designed to streamline code reviews for development teams. The tool connects to GitHub repositories and uses AI to analyze code changes within a secure, ephemeral sandbox environment. It aims to assist human reviewers by identifying potential bugs and providing context, inspired by productivity tools like Linear and Superhuman. AI

ActorCore, an open-source framework for AI agents, has been released, offering stateful serverless execution that aims to be significantly cheaper than existing sandbox solutions. It leverages WebAssembly and V8 isolates for near-zero cold starts and can be deployed across various platforms. The framework supports multiple AI models and provides granular security controls, with options for self-hosting or using a managed cloud service. AI

The blog post "Python at the Speed of Rust" introduces a new approach to Python performance by leveraging Rust. It details how to integrate Rust code into Python projects, aiming to achieve significant speedups for computationally intensive tasks. The author demonstrates practical methods for this integration, offering a way to enhance existing Python applications without a complete rewrite. AI

Hugging Face and Cloudflare have announced a partnership to integrate Hugging Face's FastRTC technology with Cloudflare's network. This collaboration aims to enhance real-time communication applications by improving the performance and scalability of speech and video streaming. The integration is expected to provide developers with more robust tools for building seamless interactive experiences. AI

Researchers have developed SeedLM, a novel post-training compression technique for large language models that utilizes pseudo-random generator seeds to encode model weights. This method aims to reduce the high runtime costs associated with LLMs by generating weight matrices on-the-fly during inference, thereby decreasing memory access and improving speed for memory-bound tasks. SeedLM achieves this by trading compute for fewer memory accesses and notably does not require calibration data, generalizing well across diverse tasks and maintaining accuracy comparable to FP16 baselines even at significant compression levels. AI

A developer is creating a versatile OCR pipeline designed to extract structured data from complex educational materials for machine learning training. The system, which supports multilingual text, mathematical formulas, tables, and diagrams, aims to achieve over 90-95% accuracy on academic datasets. It generates AI-ready outputs in JSON or Markdown, including semantic annotations for visual content, and is built using various tools like Google Vision API and OpenAI API. The project's public release has been delayed due to the developer's academic commitments but is expected once the system is finalized. AI

Gradio, a popular open-source Python library for building machine learning interfaces, has surpassed one million users. The platform facilitates the creation of web UIs for AI models, enabling developers to easily share and demo their work. This milestone highlights the growing demand for accessible tools in the AI development community. AI

Hatchet, a new task orchestration platform, has been released, offering a robust solution for managing background tasks, AI agents, and durable workflows at scale. Built with a unique approach using Postgres as its durability layer, Hatchet aims to simplify self-hosting while providing features like automatic retries, real-time monitoring, and multi-language support. The platform is available as a cloud service or for self-hosting, targeting applications where reliability and scalability are critical. AI

Hugging Face has detailed its approach to managing sensitive information like API keys and credentials across its AI infrastructure. The company implemented a robust secrets management system to ensure security and compliance as its operations grew. This system allows for secure storage, distribution, and rotation of secrets, which is crucial for maintaining the integrity of AI models and services. AI

Daniel from Zep has developed an integration for the Cursor IDE that provides persistent memory across coding sessions. This system uses Zep's open-source Graphiti framework and its Model Context Protocol (MCP) to store and retrieve user preferences, project specifications, and coding standards. The goal is to enhance the AI-assisted IDE by allowing it to remember crucial context without constant user input, adapting in real-time to changes in frameworks or standards. AI