Why it matters: Task-scoped privileges for AI agents address a core risk as agents gain infrastructure access.
Cisco's Privileged Infrastructure Access aims to scope both administrator and AI agent privileges to the specific task at hand, SiliconANGLE reports.
AIOZ Network's AIOZ AI unit has launched an MCP Server and Agent Skills, letting AI agents connect to its decentralized infrastructure.
Why it matters: Shows merchants starting to actively block agentic shopping on their sites.
Secondhand luxury retailer The RealReal updated its bot defense system to make it harder for AI agents to shop its site, days after a report showed an agent exploiting item reserves.
iProov released a human authorization service compliant with the HAPS standard, designed to gatekeep AI agent gateways by verifying human approval for agent actions.
Why it matters: A former GitHub CEO building an agent-first rival signals the coding platform is being rearchitected around AI agents.
Entire is a new GitHub competitor built for humans and AI coding agents, launched by former GitHub CEO Thomas Dohmke. The platform is live now and organizes work around agents.
Tairex, incubated by Tsinghua's Institute for AI Industry Research, unveiled its Multidisciplinary Human-Agent Collaboration ecosystem for clinical decision support and care team coordination.
Google Workspace now opens, previews and edits .md files natively in Google Docs and Drive, so developers and AI agents can collaborate on documentation without converting files.
Google's new AI Quest, Polar Peaks, is an interactive lesson that teaches students how AI models are tested and evaluated, using a cartoon planet scene with characters.
Why it matters: A quarter of Microsoft's workforce eligible for buyouts signals how AI spending is reshaping big tech staffing.
Microsoft is offering voluntary buyouts to up to 60,000 employees across the US, UK, Canada and India, giving staff financial support and career planning resources as it retools its workforce around AI.
Microsoft is adding an AI-powered security feature to Teams that analyzes conversations to detect phishing, scams and identity theft, rather than only checking links and files.
REA from Adafruit lets a coding agent inspect software, read its code and watch it run to explain or rebuild features; it installs with npx rea-agents@latest setup.
Nace AI's Drex for Search applies its Drex decision models to document search for AI agents. It reports 92.05% on BrowseComp-Plus with 12.6 search calls and fewer agent turns on OfficeQA Pro.
Konnect is a native KiCad 10 plugin that lets Claude or another LLM work on schematics and boards through the Model Context Protocol. It ships as a single Rust binary of 20-25 MB installed from a roughly 9 MB download.
Btab launched Btab.app, an AI-powered platform intended to connect manufacturers and entrepreneurs through its global commerce ecosystem.
Tipalti, which calls itself an agentic AI platform for finance, launched Tipalti Treasury, a fully integrated treasury solution built on its 2025 acquisition of Statement.
B2Prime Group launched a connector in Claude's directory, letting users reach its market intelligence and trading analytics through the AI assistant.
Erol Sen created a system that prints a two-page newspaper every morning. A Claude routine writes the issue in the cloud, GitHub Actions lays it out, and a Raspberry Pi Pico 2 W sends it to a home printer.
A free batch ComfyUI workflow chains Qwen Image 2.1, MiniMax H3, Qwen3-TTS and RTX upscaling to produce short history documentaries locally, with a tutorial linked.
Why it matters: Inline prompt inspection aims to stop sensitive data leaking through enterprise LLM use.
Fortinet expanded its AI security lineup with FortiGuard Inline AI Security, a service that scans AI prompts and blocks data leaks without extra security systems. Check Point separately launched firewall-as-a-subscription protection.
SmartKargo released a portfolio of AI apps aimed at helping airlines reduce risk, drive revenue and cut costs.
Rentify launches Ally, described as the UAE's first standalone AI agent for property maintenance, handling maintenance requests for properties.
Why it matters: Embedding agents directly into enterprise ERP setup could cut implementation costs and time.
Oracle introduced AI agent-ready setup and configuration for Fusion Applications to speed implementations, plus Private Agent Factory 26.10 with Deep NL2SQL, as NetSuite added new agents.
Rightboat debuted an AI search plugin designed to surface its boat listings in ChatGPT and drive leads, an early example of AI commerce media for a niche marketplace.
iMerit, the data annotation and AI training services firm acquired by ExlService in July, launched Scholars, a platform helping AI developers recruit and manage domain experts to evaluate, fine-tune and validate models.
Chargebacks911 launched an AI agent purchase indemnification product aimed at shielding merchants from chargeback risks arising from agentic commerce purchases made by AI shopping agents.
Why it matters: One of the largest equity raises ever to fund AI infrastructure signals the scale of capex now required.
Meta launched a $35B stock offering as part of a raise of up to $95B in equity to fund its AI expansion, with Nvidia, AMD and Broadcom reportedly weighing investments.
ADP has launched new AI tools, with Automatic Data Processing stock awaiting its results, ad-hoc-news.de reports.
European open-source specialist Aiven is pitching an LLM gateway aimed at reducing token costs, hoping to do for LLMs what it does for databases.
Podigee added AI-assisted podcast publishing that works inside ChatGPT and Claude, letting users create and manage podcast content via assistants.
Sapiens launched Sapiens Brain, a knowledge graph built from its 40 years of insurance data, to speed delivery of agentic AI solutions to customers.
Ollama adds Phonon speech generation support on Apple Silicon Macs, letting users run local text-to-speech alongside its local LLM runner.
Marissa Mayer has built an AI assistant that learns from a user's camera roll to offer personalized help and recall, per The Verge.
Why it matters: An open real-time speech-to-speech translator pushes simultaneous interpretation into the open-weights ecosystem.
Hy-SimulMT is a real-time simultaneous speech-to-speech translation model from Tencent Hunyuan for bidirectional Chinese-English, with GGUF and interleaved GGUF builds on Hugging Face.
Why it matters: Extending endpoint security to AI agents addresses a fast-growing gap as autonomous agents run on enterprise machines.
Palo Alto Networks is extending its endpoint protection platform to cover AI agents, aiming to secure autonomous agent activity alongside traditional endpoints.
Video editor CapCut is launching inside ChatGPT, letting users create and edit videos directly in the chat, per SoyaCincau.
Honeywell Technologies launched Forge Plant Cognition, an industrial AI platform built with Nvidia and Dell, deployed at Dangote's Nigeria refinery to improve productivity and cut costs.
Why it matters: Open cache-engineering server lowers inference cost for LLM agents.
AKV is an SGLang-based serving system that lets LLM agents drop and reposition KV cache instead of summarizing history; Tsinghua researchers say it cuts average cost 27.7% across four agent benchmarks.
Why it matters: Extreme quantization lets a 30B MoE with huge context run on consumer GPUs.
A ternary (1.58-bit) 30B-A3B MoE model, quantized to GGUF, runs on a 12GB GPU with 368K context, 172 tok/s decode and 7000+ tok/s prefill in both stock llama.cpp and the PrismML fork. Nvidia also released a Nemotron-Personas-Germany dataset.
Why it matters: India's regulatory framework will shape compliance for one of the world's largest AI markets.
IT Minister Ashwini Vaishnaw said India's draft AI regulations could be out in November as the country's consultation paper nears release; separately, Pakistan joined China's WAICO bloc.
openskycam is an open-source 4K sky camera setup that Claude Code can operate to detect and sort meteors; its builder says it caught 17 meteors in one Saturday night.
Orbbec's Gemini 330 series are active stereo cameras for robotics and automation that compute depth on a proprietary MX6800 ASIC, offloading the host computer.
Subtle opened orders for Voicebuds, $249 earbuds that let people dictate and use AI quietly, claiming fewer transcription errors; the product page still labels the earbuds an active beta.
The BVIO Violoop is a hardware device that acts as an AI agent for your PC, per CoolThings.com; details on pricing and availability were not given.
Why it matters: One of the largest compute commitments ever, reshaping AI infrastructure.
Anthropic revealed it will pay SpaceX up to $120.5 billion for AI compute, as part of an Apollo-led $40 billion GPU lease deal with a binding SpaceX pledge. Anthropic is leasing data centers to reduce reliance on Amazon and Google.
Agile One SN is a humanoid robot platform aimed at research use, per All-About-Industries.
Why it matters: Software to keep massive GPU training clusters resilient and productive.
Nvidia Mission Control 2.5 adds autonomous recovery to Kubernetes for large GPU clusters. Nvidia Minerva brings AIOps to AI factories through Dell, Lenovo and Supermicro, turning telemetry into peak GPU yield.
Why it matters: Apple's tablet and OS updates expand on-device AI hardware and Siri features.
Apple's new OLED iPad mini ships with A20 Pro; the 512GB model gets a 6-core CPU and dual 16-core Neural Engine, while 128GB and 256GB get a 5-core CPU. macOS 27.2 and watchOS 27.2 add Siri AI features.
Why it matters: A foundation model for humanoid whole-body control from a major Chinese tech player.
Amap and Alibaba Token Hub introduced λ0, a foundation model teaching humanoid robots coordinated walking and manipulation by learning from first-person video synced with body and hand movements.
DT Research unveiled the LT365PA, a rugged laptop with NVIDIA graphics and AI processing aimed at mission-critical field teams.
Why it matters: One of the largest robotics rounds shows physical AI construction is drawing major capital.
Bedrock Robotics, which makes driverless excavators and other autonomous construction machines already on paid job sites, raised $600M at a $4.1B valuation to scale across the US.
Why it matters: Puts AI models for EDA directly inside chipmakers' own facilities, shifting chip design automation.
Cadence launched a suite of Nvidia-backed AI models for chip design that semiconductor companies can run in-house, alongside a sovereign AI data center in Munich to support electronic design automation workflows.
Realbotix launched its Echo humanoid robot line, expanding its lineup of companion-style humanoid robots.
Anjali Sardana, founder of Pronto, has set up Provision Labs, a physical AI venture based in the US, according to Inc42.
Aimoga, a unit tied to automaker Chery, opened global preorders for a home humanoid robot, marking the carmaker's move beyond vehicles.
Why it matters: Shows AI startups moving into autonomous weapons and drone defence in Europe.
German military AI startup Helsing unveiled Contra, an AI-controlled autonomous air defence system against scaled drone and missile attacks, plus two interceptor missiles including AD-1 I.
Why it matters: Ties humanoids directly into industrial machining workflows.
Mitsubishi Electric developed a machining system that integrates humanoid robots with CNC machine tools, which it calls an industry first, for automated machining tasks.
Why it matters: A major fiber build aimed at linking India's AI data centers and supporting national AI capacity.
Tata Communications plans to invest up to $450 million in a long-haul fiber optic network linking India's key AI hubs, connecting AI data centers to build out the country's AI infrastructure.
Hyundai launched the Bayon in India at ₹9.99 lakh, featuring Level 2 ADAS and a generative AI assistant among its tech features.
Why it matters: A new neocloud with gigawatts of power and a billion-dollar Nvidia order adds capacity.
SWI Compute emerged from stealth to turn gigawatts of power into Nvidia AI factories, starting in Europe. It has ordered about $1B in Nvidia systems and secured 3.6 GW of power access for a planned UK buildout.
South Korean entertainment tech startup Galaxy Corp. launches an arena in Dubai for robot performances, per Yonhap.
A Xiaomi technology committee team published HachimoDock, a ~150 yuan desktop AI device that routes voice tasks to PC agents and controls Xiaomi smart home devices.
Why it matters: A startup claims a tenfold cut in robot action-chunk inference latency, key for embodied AI.
Three-month-old Shenzhen startup Primotion trained its own embodied model, Paimon-0, which it says tops leaderboards and cuts action-chunk inference tenfold to 300 ms.
Why it matters: A large industrial supplier committing $50M signals growing institutional capital for physical AI and robotics startups.
Misumi Americas launched Misumi Ventures, a $50 million early-stage fund for robotics, automation, advanced hardware, physical AI and industrial software, targeting 20-30 seed and early-stage companies.
beRobox introduced 3D Dynamic Safety, a module for its PALTZ robotic palletizers that moves safety detection with the robot instead of keeping it fixed around the machine.
Striding AI introduced retail robots and a management platform at APRCE 2026.
Apple's iOS 27, released September 14 after a June WWDC unveiling, focuses on a rebuilt Siri AI with its own app, among other features still rolling out in betas.
PCMag reviews Clearafi, a tool that fact-checks AI answers across multiple models, priced at $50.
Dyad is an AI-powered app builder. Version 1.21.0 is out for macOS, Windows and Linux.
AnythingLLM is a private desktop app for chatting with AI models and your own documents locally. Version 1.19.0 is out for macOS, Windows and Linux.
Microsoft Copilot analytics now break out usage by internal skills, agents and MCP servers, giving admins visibility into which agentic tools employees use.
An experimental Apache-2.0 4B model built on Qwen3.5-4B for general coding with a focus on generating interactive Three.js 3D scenes; an early preview still in training.
Google is building a home for everything users have asked Gemini to remember, so they may no longer need to explicitly ask Gemini what it remembers.
Databricks says its AI Search now natively indexes Streaming Tables, Materialized Views and Managed tables using Iceberg, and its October 21 AI Runtime GA brings managed GPUs into training workflows.
AIOZ Network's AIOZ AI unit has launched an MCP Server and Agent Skills, letting AI agents connect to its decentralized infrastructure.
AWS post walks through a SageMaker SDK v3 reference architecture that trains, quality-gates and compiles per-camera models with SageMaker Neo and versions thousands of them across a fleet.
Supabase's Paul Copplestone and Ant Wilson discuss what AI agents building more software means for backend workloads, agent-friendly schemas and running long-lived jobs alongside the database.
Deepscient is a JavaScript GitHub project for AI-human collaboration on scientific problems. It organizes work into 12 stages, 55 gates, 36 tools and 98 skills with evidence chains and reproducibility. It gained 100 stars in its first week.
AWS tutorial shows building an agent that routes across two Amazon Bedrock knowledge bases, retrieves via AgentCore Gateway and returns cited answers, with seven observability layers and on-demand and continuous evaluation.
MixVio AI expanded its creative catalog to 28 model families and 53 generation routes for creating AI media.
AOL.com offers troubleshooting advice for users whose Google Gemini chat history is disappearing from the app.
Why it matters: A former GitHub CEO building an agent-first rival signals the coding platform is being rearchitected around AI agents.
Entire is a new GitHub competitor built for humans and AI coding agents, launched by former GitHub CEO Thomas Dohmke. The platform is live now and organizes work around agents.
Anthropic's full Claude API is now usable via Google Cloud billing and account permissions, letting enterprises pay and govern access through their existing cloud accounts.
Google Workspace now opens, previews and edits .md files natively in Google Docs and Drive, so developers and AI agents can collaborate on documentation without converting files.
A developer used Claude to write most of the C# code for a Valheim mod that turns the game into a city builder, with vikings paying taxes and boss camps, reaching #1 on r/valheim.
Confident AI is a hosted platform, built on the open-source DeepEval framework, that converts failures of production AI agents into tests for the next release. Customer Finom reports faster iteration, though the case study does not measure AI quality.
Transformers v5.21 adds new models including MiniCPM-OCR, a document-understanding model from OpenBMB's MiniCPM-V family that keeps the MiniCPM-V 4.6 architecture.
Because big platforms offer little human support, users ask ChatGPT for tech help, and it increasingly points them to 404 Media as a tech support source for major corporations, per 404 Media.
REA from Adafruit lets a coding agent inspect software, read its code and watch it run to explain or rebuild features; it installs with npx rea-agents@latest setup.
Nace AI's Drex for Search applies its Drex decision models to document search for AI agents. It reports 92.05% on BrowseComp-Plus with 12.6 search calls and fewer agent turns on OfficeQA Pro.
Konnect is a native KiCad 10 plugin that lets Claude or another LLM work on schematics and boards through the Model Context Protocol. It ships as a single Rust binary of 20-25 MB installed from a roughly 9 MB download.
36Kr reports that open-source inference frameworks vLLM and SGLang have surpassed DeepSeek's own performance, intensifying competition over LLM inference speed.
Android Police walks through using Gemini in Google Sheets to track a personal budget, saying it saved hours of manual work.
Erol Sen created a system that prints a two-page newspaper every morning. A Claude routine writes the issue in the cloud, GitHub Actions lays it out, and a Raspberry Pi Pico 2 W sends it to a home printer.
A tutorial showing developers how to build a first AI agent in Python using OpenAI's Agents SDK, walking through setup and core concepts.
A how-to on disabling Google's personalized AI features and stopping the company from using personal emails and photos for targeted ads.
An experimental Apache-2.0 4B model built on Qwen3.5-4B for general coding with a focus on generating interactive Three.js 3D scenes; an early preview still in training.
Transformers v5.21 adds new models including MiniCPM-OCR, a document-understanding model from OpenBMB's MiniCPM-V family that keeps the MiniCPM-V 4.6 architecture.
d3-turbo is a 27B multimodal decision model from vLLM Semantic Router. It reads text, JSON, images or video once and answers by picking options, yes/no, rating on a scale, or locating objects.
A ternary (1.58-bit) 30B-A3B MoE model, quantized to GGUF, runs on a 12GB GPU with 368K context, 172 tok/s decode and 7000+ tok/s prefill in both stock llama.cpp and the PrismML fork. Nvidia also released a Nemotron-Personas-Germany dataset.
SharpaDex v1.0 is a multimodal dataset of real-world teleoperated demonstrations for bimanual dexterous manipulation, covering 56 tasks like assembly and tool use.
Cline CLI 3.0.74 is an autonomous coding agent CLI that creates and edits files, runs commands and uses the browser, installed via npm. Cline for VS Code 4.1.26 and Kilo Code CLI 7.8.18 also updated.
SimReady Central USD Search is a demo app for searching 3D assets. Nvidia's Hugging Face Space runs OpenSearch, Neo4j, Redis and SigLIP2 embeddings on an L4 GPU.
Nanbeige4.3-33B is an open-weight model on Hugging Face for agentic coding and office work, with a 512K-token context; its maker plans to distill it into a 3B version.
Deepscient is a JavaScript GitHub project for AI-human collaboration on scientific problems. It organizes work into 12 stages, 55 gates, 36 tools and 98 skills with evidence chains and reproducibility. It gained 100 stars in its first week.
Transformers v5.21 adds new models including MiniCPM-OCR, a document-understanding model from OpenBMB's MiniCPM-V family that keeps the MiniCPM-V 4.6 architecture.
A browser extension for Chrome, Edge and Firefox identifies and removes Bilibili commenters run by DeepSeek models or AI agents, using offline rules plus the DeepSeek API for confirmation.
A JavaScript plugin for the DSH tool that retries requests at token level and reshapes prefill, refusals and finish reasons without patching the host. 188 tests included.
Canvas-MCP is a read-only MCP server in JavaScript that lets Claude, ChatGPT and other AI assistants read data from the Canvas learning management system; it drew 102 stars in its first week.
CVEngine is an open-source Python tool for AI-first external attack surface assessment, with continuous discovery, evidence-based vulnerability validation and MCP integration for developer agents.
An open-source JavaScript tool that takes a manga original from script and image generation through layout to PDF output, aimed at Codex users; the repo hit 100 stars in its first week.
Apple's iOS 27, released September 14 after a June WWDC unveiling, focuses on a rebuilt Siri AI with its own app, among other features still rolling out in betas.
PCMag reviews Clearafi, a tool that fact-checks AI answers across multiple models, priced at $50.
AnythingLLM is a private desktop app for chatting with AI models and your own documents locally. Version 1.19.0 is out for macOS, Windows and Linux.
Google is building a home for everything users have asked Gemini to remember, so they may no longer need to explicitly ask Gemini what it remembers.
MixVio AI expanded its creative catalog to 28 model families and 53 generation routes for creating AI media.
AOL.com offers troubleshooting advice for users whose Google Gemini chat history is disappearing from the app.
A developer used Claude to write most of the C# code for a Valheim mod that turns the game into a city builder, with vikings paying taxes and boss camps, reaching #1 on r/valheim.
Because big platforms offer little human support, users ask ChatGPT for tech help, and it increasingly points them to 404 Media as a tech support source for major corporations, per 404 Media.
Android Police walks through using Gemini in Google Sheets to track a personal budget, saying it saved hours of manual work.
Erol Sen created a system that prints a two-page newspaper every morning. A Claude routine writes the issue in the cloud, GitHub Actions lays it out, and a Raspberry Pi Pico 2 W sends it to a home printer.
A how-to on disabling Google's personalized AI features and stopping the company from using personal emails and photos for targeted ads.
Game Your Game launched an Explore Golf Course feature for its GameGolf KZN AI app, giving golfers course insight and social play.
Rentify launches Ally, described as the UAE's first standalone AI agent for property maintenance, handling maintenance requests for properties.
shattered.io compares GPT Image 2.5 with three competing AI image generators to pick the best option in 2026.
Android Police asked Gemini to design a 3D-printed part to cat-proof a door lock; the AI's design nearly worked.
Podigee added AI-assisted podcast publishing that works inside ChatGPT and Claude, letting users create and manage podcast content via assistants.
Marissa Mayer has built an AI assistant that learns from a user's camera roll to offer personalized help and recall, per The Verge.
A Maeil Business column suggests using AI to organize subscribed services and cut off unwanted ones automatically.
Review compares Meta AI's features and pricing with ChatGPT, assessing how the assistant stacks up across everyday tasks.
SurgePix is an AI image tool that lets you edit any layer, swap text and repaint regions while keeping a brand kit; open-source agent skills let Claude Code and Cursor generate and edit visuals.
Kiwi is a proactive AI assistant that tracks trips, plans, bills and people, texts you at the right moment and can act on any app or website once you approve.
Google Messages may soon use a Gemini AI feature to flag scam text messages before users fall for them, according to a report.
LG Uplus launched U+tv Today, a dashboard on the TV's first screen combining weather, news, finance and local info without extra apps or logins, and redesigned the U+tv home screen around it.
Elon Musk says Grok Bot will call businesses on a user's behalf and report back results, joining Meta's Muse among AI agents that make calls for users.
Meshy, which generates 3D models and assets from prompts, opens more free browser-based tools for AI-powered 3D creation, broadening access beyond paid tiers.
Dyad is an AI-powered app builder. Version 1.21.0 is out for macOS, Windows and Linux.
Microsoft Copilot analytics now break out usage by internal skills, agents and MCP servers, giving admins visibility into which agentic tools employees use.
Databricks says its AI Search now natively indexes Streaming Tables, Materialized Views and Managed tables using Iceberg, and its October 21 AI Runtime GA brings managed GPUs into training workflows.
AIOZ Network's AIOZ AI unit has launched an MCP Server and Agent Skills, letting AI agents connect to its decentralized infrastructure.
AWS Machine Learning · 2h
#
AWS post walks through a SageMaker SDK v3 reference architecture that trains, quality-gates and compiles per-camera models with SageMaker Neo and versions thousands of them across a fleet.
Supabase's Paul Copplestone and Ant Wilson discuss what AI agents building more software means for backend workloads, agent-friendly schemas and running long-lived jobs alongside the database.
AWS Machine Learning · 3h · 2 sources
#
AWS tutorial shows building an agent that routes across two Amazon Bedrock knowledge bases, retrieves via AgentCore Gateway and returns cited answers, with seven observability layers and on-demand and continuous evaluation.
Entire is a new GitHub competitor built for humans and AI coding agents, launched by former GitHub CEO Thomas Dohmke. The platform is live now and organizes work around agents.
Anthropic's full Claude API is now usable via Google Cloud billing and account permissions, letting enterprises pay and govern access through their existing cloud accounts.
Google Workspace now opens, previews and edits .md files natively in Google Docs and Drive, so developers and AI agents can collaborate on documentation without converting files.
Confident AI is a hosted platform, built on the open-source DeepEval framework, that converts failures of production AI agents into tests for the next release. Customer Finom reports faster iteration, though the case study does not measure AI quality.
REA from Adafruit lets a coding agent inspect software, read its code and watch it run to explain or rebuild features; it installs with npx rea-agents@latest setup.
Nace AI's Drex for Search applies its Drex decision models to document search for AI agents. It reports 92.05% on BrowseComp-Plus with 12.6 search calls and fewer agent turns on OfficeQA Pro.
Konnect is a native KiCad 10 plugin that lets Claude or another LLM work on schematics and boards through the Model Context Protocol. It ships as a single Rust binary of 20-25 MB installed from a roughly 9 MB download.
36Kr reports that open-source inference frameworks vLLM and SGLang have surpassed DeepSeek's own performance, intensifying competition over LLM inference speed.
A tutorial showing developers how to build a first AI agent in Python using OpenAI's Agents SDK, walking through setup and core concepts.
Amp is a frontier coding agent and development environment; these releases update its CLI, installable via npm with @ampcode/cli, for running the agent from the terminal.
Vibe code audits are emerging as a quality gate for AI-generated software, reviewing code produced by AI coding tools before it ships.
Unite.AI looks at developers using AI tools to clone paid software such as Adobe, weighing accusations of theft against enthusiasm for free open-source alternatives.
Oracle introduced AI agent-ready setup and configuration for Fusion Applications to speed implementations, plus Private Agent Factory 26.10 with Deep NL2SQL, as NetSuite added new agents.
An explainer on computer-using agents, which operate software via screens, cursors and keyboards like people, and the visual uncertainty, long task horizons and broad permissions that require careful controls.
AKV is an SGLang-based serving system that lets LLM agents drop and reposition KV cache instead of summarizing history; Tsinghua researchers say it cuts average cost 27.7% across four agent benchmarks.
openskycam is an open-source 4K sky camera setup that Claude Code can operate to detect and sort meteors; its builder says it caught 17 meteors in one Saturday night.
N8N Chat UI is a no-code builder for branded AI chat widgets for n8n users, agencies and SaaS teams; it connects to any n8n workflow and adds streaming, voice, file uploads and a JavaScript API.
AI Bridge is a Mac app that runs Claude Code, Codex, Grok and OpenCode side by side, each real CLI in a real terminal on your own subscription, organized as teams alongside cluster and database chats.