OpenAI rolled out GIF generation in ChatGPT on November 4 as part of its Images 2.6 update: users describe a scene or upload a photo and ChatGPT animates it into a looping GIF.
An Irish startup uses AI to convert a single product photo into a working online store in about three minutes, according to The Irish Times.
Why it matters: Apple's first China AI rollout leans on a local model partner, a template for regulated markets.
Apple's iOS 27.3 developer beta starts testing China-specific Apple Intelligence, with Alibaba's Qwen model integrated into Siri; the full release is expected in December.
Why it matters: Raises safety concerns about recursive self-improvement in frontier AI systems.
Anthropic warns that AI could soon start improving itself at a 'dramatically accelerated' pace, with one expert calling 2026 the 'last normal year.'
Why it matters: A major robotics M&A deal is stopped by Chinese regulators, shaping global robotics consolidation.
China blocked SoftBank's $5.4 billion acquisition of ABB's robotics division, which focuses on the Chinese market and has a mega-factory in Pudong.
Why it matters: Adds an open-weight reasoning model tuned for embodied and physical-AI tasks.
An embodied reasoning model for robots and physical tasks with 35B total and 3B active parameters. China Merchants' lab says a capability-driven data engine let it beat its base model on ten benchmarks.
A swarm architecture that gives persistent AI agents their own identities and inboxes, letting them join Feishu and DingTalk groups as digital employees via WorkSwarm.
The openPangu-2.1-Pro model is live first in Huawei's Xiaoyi Work agent, labeled HarmonyOS task enhancement, days after openPangu-2.0 was set to leave Huawei Cloud MaaS.
FlutterFlow launched an AI-native builder that generates iOS and Android apps from prompts, aimed at letting users go from idea to working mobile app.
Fingerprint extended its bot detection to uncover concealed traffic from bots like Muse, Instinct and Grok, as AI agents increasingly hide their identities online.
HS Ad integrated Netflix ad reach prediction into its AI agent DASH, letting advertisers forecast and analyze Netflix ad effectiveness in advance.
Why it matters: Signals formal US rules governing how AI labs operate with the military.
The Pentagon's chief technology officer said the US government will lay out this month which laws apply to AI labs, according to Reuters.
Why it matters: A $5B raise would rank among the largest AI-biotech financings and signal investor appetite for AI drug discovery.
Alphabet's Isomorphic Labs is negotiating roughly $5 billion in new funding to support its AI-driven drug discovery work, according to people familiar with the matter.
Why it matters: A small 8.8B model matching far larger systems on math and coding signals gains in post-training efficiency.
webAI released TwIL-8B, an 8.8B reasoning model post-trained from IBM Granite 4.2 8B. It scores 98.2 on MATH-500 and 81.6 on LiveCodeBench but trails peers on knowledge benchmarks.
Certinia is adding AI to the software professional services firms use to sell projects, assign staff, deliver work and bill clients, targeting billable-hours delivery.
From ex-Life360 directors Charly Delaroche and Marc McCole, Anywhere is a social map app in Y Combinator's Fall 2026 batch where AI agents can read, edit and undo changes to shared maps.
A free AI skill called /fuck-cancer organizes cancer records into a five-part caregiver brief, with explicit limits against diagnosis or treatment advice.
Stability AI's November 2026 release notes for Stable Audio describe updates to its text-to-audio music generation tool. The post covers what changed in the latest version.
BeyondTrust launched AI agent security capabilities, released with accompanying research, aimed at securing enterprise AI agents.
Forter introduced new capabilities built on an agentic commerce engine, per Retail TouchPoints, extending its fraud and trust platform to AI shopping agents.
MariaDB released Enterprise Platform 2027, generally available, combining transactional, analytics, in-memory and AI capabilities with agent memory, retrieval and response caching; peak throughput up 56%.
Why it matters: Meta buying a multi-agent coding startup signals consolidation in AI developer tooling.
Conductor lets software engineers deploy teams of agents to work on different parts of a project simultaneously; Meta has acquired the startup.
Why it matters: Signals Apple tightening internal AI spend while rivals ramp up, a notable cost-control move.
Apple is reducing employee budgets for AI tool usage by roughly a third in a cost-control push, even as it keeps AI central to its operations, Bloomberg reports.
A free ComfyUI workflow called RefMods lets users extract a character and voice reference once and reuse them across MiniMax H3 clips, running on a 6 GB GPU.
BotGauge launched an AI agent red-teaming platform with over 200 attack strategies, plus tracing and evaluation tools for testing agent security.
Enrich Labs' Agent Conversion Optimization tracks whether shopping agents recommend your brand, monitoring Muse, Instinct and Dots daily from $199 a month, capped at 100 customers.
Why it matters: A 65B open-weight model claiming US SOTA in agentic coding narrows the gap with larger models.
Ripple-65B Preview is a 65B dense open-weight model for agentic coding, trained on 12 trillion tokens with up to 1M context and 80.5 on Terminal-Bench 2.1, claimed as US open-weight SOTA.
Why it matters: Enterprise routing layer aims to cut coding agent costs while holding quality.
Grid is a routing, gateway and guardrails layer that lets enterprises control AI coding agent traffic and spend. Baseten opened a public preview; its Router reported nearly 70% cost-per-task savings with a 0.9-point quality tradeoff.
Xiora Technologies introduced Xiora AI, a web platform offering personalized AI video companions that users can interact with online.
Why it matters: Cheap fast video generation broadens access for pro film and TV production pipelines.
Kuaishou's Kling 4.0 Flash video model is generally available, priced at $0.094 per second with 720p output, and is integrated into professional film and TV workflows.
Why it matters: First clear IPO timeline and enterprise revenue disclosure for the leading AI lab.
Greg Brockman said OpenAI is ready to IPO in 2027 and now draws most of its sales from enterprises, while policy chief Chris Lehane pushes AI's role in the 2028 election.
Jobber lets eligible US home and commercial service businesses be discovered and booked through Google Search's AI Mode, with requests flowing into existing Jobber workflows and no lead fees.
Contextbase stores AI agent context — semantics, knowledge, memory, traces and evaluations — inside PhoenixAI's database. It is in private preview; ObserveStack is the first app built on it.
Suno 1.96.0 lets users upload MIDI files in Create, giving more control over melodies and structure when generating AI songs.
Phone Harness gives AI agents control of a real iPhone or a rented cloud phone. Founder Shawn Pana open-sourced the harness for personal devices and sells persistent cloud phones.
Why it matters: Makes multi-agent coding in a single editor session mainstream for developers.
The new VS Code release lets developers run Claude, Codex and GitHub Copilot agents within the same agent session, per Windows Report.
The Dharma Field Extractor is a 2B model that takes any document and outputs structured fields as JSON. Dharma-AI's Gabriel Pimenta released it on Hugging Face.
Unsloth released Dynamic v3.1 quantized GGUFs for Qwen3.8-Flash-Next, claiming about 15% better accuracy than GSQ RCO at similar file size; users report IQ_3S GSQ RCO nearing BF16 quality.
AT&T's Connected Life home security system now integrates with Gemini for Home, adding Gemini-powered event summaries and video history search. AT&T calls it the first home security product to use Google Home Premium.
Viggle Remake is an open-weights model for recasting people and editing scenes in existing videos while keeping the original motion; weights are on Hugging Face and a demo app is available.
Korean fabless chipmaker Mobilint improved Ray support so the open-source distributed AI framework automatically detects its neural processing units.
CEO Carl Pei confirmed Nothing's first AI-native device will arrive in December 2026. The screenless product is pitched as a productivity tool and won't be available as a standalone app.
TECO Electric and Machinery is extending its motor and drive expertise into rotary actuators, building joint modules for humanoid and quadruped robots and says it has already shipped units.
Matrix Robotics debuted its MATRIX-3 robot at Hong Kong Fintech Week, completing a first real-world retail scenario demonstration in Hong Kong.
Why it matters: A simulation world model could speed up autonomous-driving training and validation.
Baidu's Apollo World model uses AI agents to simulate traffic and teach autonomous driving systems. It is already supporting Apollo Go's global testing and validation.
Korean startup 42Maru released Appliance42, a hardware-software bundle combining GPU, NPU and LPU infrastructure with its agentic AI solution, so organizations can deploy AI immediately without building their own compute.
oToBrite is partnering with Yuan High-Tech to integrate its 3D binocular stereo vision camera module with Nvidia Jetson Thor, creating an end-to-end vision solution for robots and unmanned vehicles.
Why it matters: Modular reference designs could accelerate how fast AI data center capacity gets built and powered on.
Nvidia's newly launched DSX POD reference design aims to shorten AI data center build times and bring capacity online faster, as Taiwanese firms race into modular AI data center integration.
PUDU displayed a semi-humanoid robot that lifts 22 lbs with 8-hour runtime, an all-terrain LiDAR robot dog and a cat-like service robot at the Robot World 2026 exhibition in Korea.
Chery's Jetour T1 SUV launched in China starting at about $12,500, with a 150 km electric range and DeepSeek AI integration in the cabin.
Curbie is a robot designed to move household trash bins to and from the curb, replacing a common chore.
Why it matters: Arm's results are a bellwether for how fast AI data-center spending reaches chip designers.
Arm beat second-quarter revenue estimates on AI and data-center demand and raised its AI chip revenue target to $2 billion, but its current-quarter forecast left investors cold.
Fern, a two-person YC startup, says three teleop episodes are enough to build a robot simulation replacing hardware evals; one customer ran 900,000 simulations.
Raven, a YC Fall 2026 startup, is taking a refundable $199 deposit toward a planned $99-$199 monthly plan for a phone built around AI agents. No shipping date announced.
Apple's updated support document says the Siri AI waitlist launched with iOS 27 is no longer used except in high-demand times; coverage also notes Siri's delayed app integration and Apple's warning on autonomous agent risks.
Why it matters: Qualcomm's pivot signals intensifying competition in AI data center silicon beyond Nvidia.
With smartphone chip sales missing estimates and shares falling, Qualcomm laid out Q4 2026 plans to expand into AI data centers via partnerships including Amazon, while pitching its Cloud AI 100 Ultra 150W accelerator.
Fermaca said it became Mexico's first Nvidia Cloud Partner and unveiled a phased AI factory platform for Mexico and Latin America, starting with Fermaca Digital Origins in Mexico City.
CUDA 13.5 adds GA support for Nvidia's Vera Rubin platform, letting developers build and run accelerated workflows on the new hardware in production.
An open-source project builds an AI-powered claw suspended from cables near the ceiling, letting it pick things up around a room instead of using a humanoid robot.
Broadcom added AI, security and recovery capabilities to VMware Cloud Foundation, its private cloud platform.
Weebit Nano launched a research program giving academic and industry researchers ReRAM hardware kits built around a 32x32 crossbar to test running AI calculations where data is stored, rather than moving it.
Why it matters: Shows scale of debt financing now flowing into AI compute and chip buildouts.
Goldman Sachs and Apollo are seeking buyers for a $40 billion SpaceX AI chip financing deal, as the big four hyperscalers reportedly seek $131 billion for AI infrastructure.
Nvidia released its new R620 Linux graphics driver series, version 620.70.06, with Wayland improvements and DLSS 5 support for games including Hogwarts Legacy and World of Warcraft.
Hyundai Motor Group demonstrates a robot that automatically refuels a hydrogen-powered cleaning truck, alongside a 5 MW hydrogen system designed to produce over 700 tons annually.
Why it matters: Shows Google and Anthropic locking in massive AI compute capacity.
TeraWulf secured a Google credit backstop for its $19 billion Anthropic data-center lease project.
An open-source TypeScript tool that restarts failed AI agent jobs from their last checkpoint instead of from scratch, keeping verified completed work and billing the remainder in USDC on Solana via x402 after you accept results.
Microsoft and Nvidia are standardizing RTX Mega Geometry technology in the new DirectX Agility SDK, giving developers a common path to the geometry tech.
OpenAI rolled out GIF generation in ChatGPT on November 4 as part of its Images 2.6 update: users describe a scene or upload a photo and ChatGPT animates it into a looping GIF.
A swarm architecture that gives persistent AI agents their own identities and inboxes, letting them join Feishu and DingTalk groups as digital employees via WorkSwarm.
FlutterFlow launched an AI-native builder that generates iOS and Android apps from prompts, aimed at letting users go from idea to working mobile app.
A Hugging Face Space demo that removes clichéd, formulaic phrasing from English and Chinese text while leaving code, URLs, numbers and quotes untouched. Runs fully in the browser with no upload or backend; the Lite build has a fixed rule list.
A small version update to the CLI for Amp, a frontier coding agent and development environment. Installable via npm i -g @ampcode/cli.
An open-source AI coding agent for VS Code that writes code from plain-language prompts, runs terminal commands and automates refactors, with planning and debugging modes. Version 7.8.26 adds the latest fixes and supports 500+ models.
Auggie is Augment Code's command-line agent that brings AI coding assistance to the terminal. Version 0.37.0 is installed via npm and runs on Mac, Windows and Linux.
OpenAI's Dots release adds inline iOS notification replies, faster calls, a spoken hang-up and more reliable Windows MCP workflows, per Samir Ahmed.
A free Windows 10/11 desktop app that turns a text prompt into a short MP4 video locally using Wan 2.2, with wide, vertical or square output and no credits or queue.
From ex-Life360 directors Charly Delaroche and Marc McCole, Anywhere is a social map app in Y Combinator's Fall 2026 batch where AI agents can read, edit and undo changes to shared maps.
Notion now syncs workspace memory so ChatGPT and Claude can share one memory via MCP, though Notion's own October 19 release notes still listed skills via MCP as coming soon.
Warp, a Rust-based terminal, shipped version 0.2026.11.03.17.08.stable_00 for macOS, Windows and Linux. It is an AI-powered command-line tool for developers.
Perplexity's Decisions API now offers optional live web searches before returning probabilities, targeting answers that change after training like a person's role or software details.
A free AI skill called /fuck-cancer organizes cancer records into a five-part caregiver brief, with explicit limits against diagnosis or treatment advice.
Raycast, a keyboard-driven launcher for controlling tools and running commands, released version 2.11.1.0 on macOS, Windows and Linux.
An open-source AI agent that brings Gemini directly into the terminal — v0.67.0 adds an IDE bridge and other changes. Free and open source for developers.
MariaDB released Enterprise Platform 2027, generally available, combining transactional, analytics, in-memory and AI capabilities with agent memory, retrieval and response caching; peak throughput up 56%.
Zed is a fast collaborative code editor; version 1.27.0 ships for macOS, Windows and Linux with its multiplayer editing features.
Why it matters: A 65B open-weight model claiming US SOTA in agentic coding narrows the gap with larger models.
Ripple-65B Preview is a 65B dense open-weight model for agentic coding, trained on 12 trillion tokens with up to 1M context and 80.5 on Terminal-Bench 2.1, claimed as US open-weight SOTA.
AWS explains connecting decision models Jev and Strands Decider to Amazon Bedrock AgentCore Evaluations via code-based evaluators, to answer narrow, repeatable agent-evaluation questions.
Why it matters: Enterprise routing layer aims to cut coding agent costs while holding quality.
Grid is a routing, gateway and guardrails layer that lets enterprises control AI coding agent traffic and spend. Baseten opened a public preview; its Router reported nearly 70% cost-per-task savings with a 0.9-point quality tradeoff.
After major Republican election losses, users asked Grok whether Elon Musk deleted his 'Vote Red Today!' post; the xAI chatbot said yes, per Yahoo. An example of AI answering questions about its owner's posts.
Contextbase stores AI agent context — semantics, knowledge, memory, traces and evaluations — inside PhoenixAI's database. It is in private preview; ObserveStack is the first app built on it.
Phone Harness gives AI agents control of a real iPhone or a rented cloud phone. Founder Shawn Pana open-sourced the harness for personal devices and sells persistent cloud phones.
Why it matters: Makes multi-agent coding in a single editor session mainstream for developers.
The new VS Code release lets developers run Claude, Codex and GitHub Copilot agents within the same agent session, per Windows Report.
Online Tech Tips walks through settings on desktop, Android and iPhone to reduce AI-generated Pins in the Pinterest feed.
AT&T's Connected Life home security system now integrates with Gemini for Home, adding Gemini-powered event summaries and video history search. AT&T calls it the first home security product to use Google Home Premium.
Replay's QA tool surfaced more than 900 bugs across 40 promoted apps, according to CEO Brian Hackett, and traced them to root causes AI agents can act on.
A Hugging Face Space demo that removes clichéd, formulaic phrasing from English and Chinese text while leaving code, URLs, numbers and quotes untouched. Runs fully in the browser with no upload or backend; the Lite build has a fixed rule list.
Ripple-65B Preview is a 65B dense open-weight model for agentic coding, trained on 12 trillion tokens with up to 1M context and 80.5 on Terminal-Bench 2.1, claimed as US open-weight SOTA.
LFM3-Omni-2B is a small multimodal model that can run entirely on-device. A Hugging Face Space demo and a React Native app using custom WebGPU kernels run it on an iPhone at about 80 tokens per second, fully offline.
Moonshot AI released Kimi K3.1 Pro, a 2.8-trillion-parameter open-weight model, with day-zero support from Huawei Ascend, AMD Instinct and vLLM/SGLang.
A model that turns a single image into a 3D asset, a fine-tune of TripoSplat, repackaged with ComfyUI-ready files under Apache-2.0. Drop-in checkpoints plug into existing TripoSplat workflows.
PentesterFlow/OffsecCyber is an Apache-2.0 English dataset of security assistant conversations covering CVE analysis, CWE references, remediation and web-appsec reasoning.
Hydrogen-1.0 is a scientific problem-solving model from Deep Principle built on GLM-5.3 that helps answer scientific questions, explain derivations and organize responses to open-ended research problems.
Idol is a Gradio demo that converts a single image into a 3D object as Gaussian splats, using DINOv3 and FLUX.2-dev image encoders; running it requires an HF token with access to those models.
A community MiniMax-H3 image model 7x smaller and 5x faster in 8 steps, plus PrunaAI's ComfyUI model and a drop-in KSampler replacement that speeds diffusion for image, edit and video models without retraining.
A Hugging Face GGUF of DeepSeek-V4.1-Flash-1017 compressed to about 1 bit per parameter, with half its 384 experts per layer removed, yielding 72.1 GB of weights for code and agent workflows.
An open-source TypeScript tool that restarts failed AI agent jobs from their last checkpoint instead of from scratch, keeping verified completed work and billing the remainder in USDC on Solana via x402 after you accept results.
A free Windows 10/11 desktop app that turns a text prompt into a short MP4 video locally using Wan 2.2, with wide, vertical or square output and no credits or queue.
An open-source AI agent that brings Gemini directly into the terminal — v0.67.0 adds an IDE bridge and other changes. Free and open source for developers.
Capri is an open-source desktop AI agent with an Edgerunners-inspired look that works with Claude Code, Codex, Antigravity and other coding agents. The C# repo drew 138 stars in its first week.
Dify is an open-source collaborative workspace for building agentic workflows and RAG pipelines with support for many AI models and tools. Version 1.19.0 updates the self-hostable platform used to move prototypes to production.
A curated GitHub list of self-hosted open-source AI tools — local LLMs, chat UIs, agents, MCP servers, RAG and speech and image generation — health-checked weekly with dead projects removed automatically.
LiteLLM is an AI gateway that lets developers call 100+ LLM APIs in OpenAI or native format with cost tracking, guardrails and load balancing. v1.108.0 signs Docker images with cosign.
agent-autoloop is a TypeScript tool that has command-line agents take on and follow up tasks in the background, pinging you with cards only when they need a decision.
ComfyUI is a modular node-graph GUI, API and backend for running diffusion image and video models locally. Version 0.44.0 is out with updates to the open-source project.
A GitHub repo provides an uncensored AI API and browser chat built on DeepSeek and GLM models, with streaming plus Python and TypeScript examples, pricing and setup guides.
Open WebUI is a user-friendly self-hosted interface for running local and API models (Ollama, OpenAI). Version 0.13.0 adds a tabbed workspace for files, terminal sessions, artifacts, previews, voice calls and comments.
OpenAI rolled out GIF generation in ChatGPT on November 4 as part of its Images 2.6 update: users describe a scene or upload a photo and ChatGPT animates it into a looping GIF.
OpenAI's Dots release adds inline iOS notification replies, faster calls, a spoken hang-up and more reliable Windows MCP workflows, per Samir Ahmed.
From ex-Life360 directors Charly Delaroche and Marc McCole, Anywhere is a social map app in Y Combinator's Fall 2026 batch where AI agents can read, edit and undo changes to shared maps.
Notion now syncs workspace memory so ChatGPT and Claude can share one memory via MCP, though Notion's own October 19 release notes still listed skills via MCP as coming soon.
A free AI skill called /fuck-cancer organizes cancer records into a five-part caregiver brief, with explicit limits against diagnosis or treatment advice.
Raycast, a keyboard-driven launcher for controlling tools and running commands, released version 2.11.1.0 on macOS, Windows and Linux.
After major Republican election losses, users asked Grok whether Elon Musk deleted his 'Vote Red Today!' post; the xAI chatbot said yes, per Yahoo. An example of AI answering questions about its owner's posts.
Online Tech Tips walks through settings on desktop, Android and iPhone to reduce AI-generated Pins in the Pinterest feed.
AT&T's Connected Life home security system now integrates with Gemini for Home, adding Gemini-powered event summaries and video history search. AT&T calls it the first home security product to use Google Home Premium.
Manus for Mac is an AI agent app that automates workflows on your own computer. Version 2.1.1 is a maintenance update for macOS, Windows and Linux.
ZDNET reports new Safari developer beta release notes point to deeper Siri AI integration and more Safari customization, plus three popular browser features.
BoltAI is an AI chat client that lets you use multiple models in one desktop app. Version 2.18.5 is a maintenance release for macOS, Windows and Linux.
At ENGAGE 2026, Smartsheet launched vsActual, a standalone AI-native planning and execution product, plus an 'execution graph' aimed at linking team tasks to organizational strategy.
A personal trainer in Wales built an AI-powered app aimed at improving gym training for users. Further product details were not specified in the report.
Google Photos is preparing Ask Photos features that could extract event details from photos, such as posters, and add them to calendars, with risks of AI errors.
MyHeritage introduces GAIA, an AI agent that helps users research their family history by searching records, building trees, and answering genealogy questions.
ChatGPT can now lock private chats with a PIN, though users should know a catch. Separately, researchers warn one ChatGPT link could have exposed private chats and hijacked accounts.
Gemini for iPhone version 1.2026.4370401 adds the Gemini 4 Argon frontier model and Gemini Omni, improving multi-step reasoning and advanced coding and app workflows in the app.
Microsoft's Copilot app for iPhone and iPad reached version 2.115.3, a maintenance update with general quality and performance improvements.
Speechify 3.23.0 is out for macOS, Windows and Linux. It is an AI-powered reading and voice assistant app.
Tesla's software update 2026.38.6 reaches wide release, bringing the Grok AI assistant out of beta and adding Latvian language support.
Prelulu is an app for immersive stories with AI characters you can text or video roleplay with. It brings real-time avatar technology to consumer scale, and users can create their own characters with live avatar capability.
Travel platform Klook introduced an AI agent aimed at streamlining operations for its merchant partners.
Version 1.24.0 of Chatbox, a desktop app for chatting with GPT-4 and GPT-3.5 through the OpenAI API, is out for macOS, Windows and Linux.
Hobbyists are using Claude to modernize classic GTA 2 graphics and to get Nintendo Switch games such as Super Mario Odyssey running natively on Apple TV, per community reports.
Microsoft and Nvidia are standardizing RTX Mega Geometry technology in the new DirectX Agility SDK, giving developers a common path to the geometry tech.
A swarm architecture that gives persistent AI agents their own identities and inboxes, letting them join Feishu and DingTalk groups as digital employees via WorkSwarm.
FlutterFlow launched an AI-native builder that generates iOS and Android apps from prompts, aimed at letting users go from idea to working mobile app.
A small version update to the CLI for Amp, a frontier coding agent and development environment. Installable via npm i -g @ampcode/cli.
An open-source AI coding agent for VS Code that writes code from plain-language prompts, runs terminal commands and automates refactors, with planning and debugging modes. Version 7.8.26 adds the latest fixes and supports 500+ models.
Auggie is Augment Code's command-line agent that brings AI coding assistance to the terminal. Version 0.37.0 is installed via npm and runs on Mac, Windows and Linux.
Warp, a Rust-based terminal, shipped version 0.2026.11.03.17.08.stable_00 for macOS, Windows and Linux. It is an AI-powered command-line tool for developers.
Perplexity's Decisions API now offers optional live web searches before returning probabilities, targeting answers that change after training like a person's role or software details.
An open-source AI agent that brings Gemini directly into the terminal — v0.67.0 adds an IDE bridge and other changes. Free and open source for developers.
MariaDB released Enterprise Platform 2027, generally available, combining transactional, analytics, in-memory and AI capabilities with agent memory, retrieval and response caching; peak throughput up 56%.
Zed is a fast collaborative code editor; version 1.27.0 ships for macOS, Windows and Linux with its multiplayer editing features.
AWS Machine Learning · 11h
#
AWS explains connecting decision models Jev and Strands Decider to Amazon Bedrock AgentCore Evaluations via code-based evaluators, to answer narrow, repeatable agent-evaluation questions.
Grid is a routing, gateway and guardrails layer that lets enterprises control AI coding agent traffic and spend. Baseten opened a public preview; its Router reported nearly 70% cost-per-task savings with a 0.9-point quality tradeoff.
Contextbase stores AI agent context — semantics, knowledge, memory, traces and evaluations — inside PhoenixAI's database. It is in private preview; ObserveStack is the first app built on it.
Phone Harness gives AI agents control of a real iPhone or a rented cloud phone. Founder Shawn Pana open-sourced the harness for personal devices and sells persistent cloud phones.
The new VS Code release lets developers run Claude, Codex and GitHub Copilot agents within the same agent session, per Windows Report.
Replay's QA tool surfaced more than 900 bugs across 40 promoted apps, according to CEO Brian Hackett, and traced them to root causes AI agents can act on.
RustingStrata is an inference engine for Qwen GGUF models written in pure Rust without ggml; CUDA kernels compile at startup via NVRTC and also run on Vulkan. It runs Qwen models from 2B to 35B-A3B.
Redis's Iris real-time context engine now offers six composable capabilities for AI agents — four generally available, two in beta — adding request routing, reusable procedures, observability and resilient messaging.
AWS Machine Learning · 14h
#
A framework for production AI agents covering architecture, prompts, file-based tools and verification agents, implemented in working code from Deep Insight, an open-source multi-agent data analysis solution.
Nvidia's NeMo tooling turns production evidence into prioritized issues and custom evaluations to improve prompts, tools, models, memory and harness code for AI agents.
datasette-llm is a plugin that adds LLM features to Datasette; version 0.1a10 fixes a bug where available models were not limited to those in the configuration.
Leidos introduced HeadWay Dispatch, software for overseeing and controlling AI agents across different vendors, aimed at organizations running multi-vendor agent fleets.
Oumi's Frontier AI Engineer agent can now build AI agents for customers and retrain the models behind them, targeting agent failures that emerge after demos.
Atlas is a free, MIT-licensed tool that creates a hidden second desktop on a Mac where computer-use AI agents can click and type while the user keeps working undisturbed.