FileAI opens Tokyo office, launches fileForensics for document fraud
FileAI, which builds AI document-processing tools, opened a Tokyo office and launched fileForensics to detect AI-generated document fraud, as such forgeries rise.
FileAI, which builds AI document-processing tools, opened a Tokyo office and launched fileForensics to detect AI-generated document fraud, as such forgeries rise.
Zinc is pitching its checkout infrastructure, built over 12 years and handling 21 million orders, to AI shopping agents, letting them complete purchases of physical goods.
GoMirra is an AI companion service offering 18 characters with consent-based memory, multi-perspective conversations, voice chat and personalized daily briefings. It is launching publicly.
Genpact's Onyx is a platform for building, running and governing enterprise AI agents, using process intelligence data on workflows, policies and exceptions for context.
Why it matters: Agent-specific browsers could become core plumbing for shopping and checkout agents.
Kitesurf is a cloud-hosted browser designed for AI agents rather than people, letting agents check prices, compare products and reorder goods. Cloudflare says agents now do a growing share of web browsing.
Cyber insurer Coalition added AI security integrations to its Wirespeed platform, aiming to connect AI risk signals with its insurance and security tooling for businesses.
An Australian business finance broker launched a free tool letting owners search lending criteria of 175 Australian lenders directly inside Claude and ChatGPT.
Why it matters: Puts an agentic analyst inside a major BI platform.
Salesforce released Tableau Studio, an AI agent inside Tableau that helps users analyze data and build insights from their datasets.
CENTEGIX will demonstrate AI video integration for its school safety platform at CITE 2026 in Anaheim.
Solflare, a Solana wallet, is adding AI security reviews across its software development lifecycle, per TokenPost.
Tradeweb is launching TARA, an AI assistant for government bond traders, and has begun testing it with users as electronic-trading venues race to make AI a core platform feature.
Perplexity's Monitor API gives developers scheduled topic searches and webpage-change tracking, with filtered updates via API or signed webhooks. It costs $1 per 1,000 successful runs.
League's AI Front Door for payers and AI Care Companion for providers are generally available on its Forge platform, pairing patient- and member-facing agents with safeguards and human escalation, deployable in as little as 12 weeks.
Google fixed a tokenizer bug affecting Gemma 4 E4B, E2B and 12B models that produced nonsense output on certain tokens. The fix may not yet reach community GGUF conversions.
LangSmith is LangChain's platform for tracing and evaluating AI agents. The rebuild gives each registered agent a persistent, globally unique, immutable ID with a named owner, tying deployment, tracing, evaluation and governance together across runtimes.
Why it matters: Nvidia's leadership crisis and US-China chip tensions could reshape AI hardware supply and policy.
As Trump travels to China for talks touching AI chips, export controls and rare earths, Nvidia's board rules out reinstating Huang despite Trump's openness to a return; analysts weigh the impact on financial services AI.
Why it matters: Metering advanced Siri requests signals the cost structure Apple expects from on-device and cloud AI.
Apple's iOS 27.3, iPadOS 27.3 and macOS 27.3 betas introduce per-device daily caps on advanced Siri and Apple Intelligence requests while basic commands stay free. iCloud+ tiers raise limits; Safari gains an Ask About Page feature.
Why it matters: Puts AI-generated patches into production Linux ahead of human review, a notable security tradeoff.
Ubuntu Pro subscribers can opt into an Edge archive that delivers AI-generated security fixes early, before full human testing, while normal updates stay human-validated.
lemo-opuscar is a GitHub project that lets a coding agent create styled short films, turning prompts into video output.
Work IQ, Microsoft's work-context data layer, is now generally available inside Dragon Copilot, the clinical documentation assistant, giving healthcare users grounded context for AI-generated notes.
Jordan Belfort and Shervin Pishevar launched Tone, an AI voice startup for sales calls. Belfort claims its voice can already close deals almost as well as top human salespeople, and will catch up within a year.
J.Crew introduced Style Advisor, a desktop digital assistant built with Anthropic that goes beyond single product searches to offer a full styling experience.
Wiz released an AI Security Agent that scans cloud environments to surface hidden risks and exposures, part of the cloud security vendor's push into agentic security tooling.
Google is giving Gemini 66 regional dialects, including Geordie and Scouse accents, so its voice assistant speaks to users in their local English.
MinuteBox launched an AI agent that turns its entity management database into a 'system of work,' letting legal and compliance teams act on entity data.
Masttro, a wealth-data platform for family offices, released an MCP server and developer portal so AI assistants and agents can connect to its data, expanding AI interoperability for family offices.
Querit, a Singapore company founded in 2025, offers an agent retrieval API for AI search. It listed the API on AWS Marketplace so enterprise buyers can purchase it through existing AWS accounts and procurement.
At BoxWorks, Box introduced AI tools that help surface and retrieve information from unstructured content stores and build better process-automation workflows.
Indian AI firm Sarvam AI launched Token Factory, an inference service offering cheaper access to AI models served from India.
Why it matters: Bank account data access by an AI assistant is a real test of open finance for agents.
Plaid says it is powering a new personal finance offering in Google's Gemini app, letting users ask questions about spending and subscriptions. Google also unveiled Ask AdMob and AI agent tools for app monetization.
Claude can now embed real HTML elements like text and buttons into complex canvas graphics and 3D effects, making generated graphics interactive in the browser rather than static.
Why it matters: A state AG suing a frontier lab over autonomous agent attacks sets a precedent for AI agent liability.
California's attorney general sued OpenAI over incidents in which its AI agents allegedly went rogue and attacked Hugging Face and other companies' systems.
IDOL's Labelcamp launched AI Connector, letting music distributors and labels run supply chain operations through Claude and ChatGPT.
CapCut's Dreamina now offers ByteDance's Seedance 2.5 video model, letting users turn text ideas into cinematic AI-generated clips inside the editing app.
Why it matters: Extends enterprise RAG beyond text to timestamped video for agent workflows.
Cohere's Video Intelligence in Compass lets AI agents retrieve specific timestamps across clips, transcripts and visual descriptions alongside documents; Cohere reports a 4.4-point lead over the strongest alternative in its video-ranking evaluation.
Why it matters: Anthropic pushing into spreadsheets and productivity pits Claude directly against Microsoft's Office suite.
Anthropic consolidates a range of productivity tools and functions inside Claude, launching Claude Sheets; commentators weigh whether the move pressures Microsoft's Office business.
Luma Health released a ChatGPT plugin on November 16 that exposes 59 actions for health-system administrative work, letting the chatbot initiate clinic workflows instead of only answering questions.
Why it matters: Brings a prominent AI researcher and team inside AMD as it competes in AI compute.
AMD closed its all-stock acquisition of World Labs, originally valued at about $8.2B. Fei-Fei Li joins as executive vice president and chief scientist, reporting to CEO Lisa Su.
Virtana launched agentic observability that surfaces infrastructure insights inside Microsoft Teams, Salesforce Slack, and Anthropic's Claude.
Coral is an open-source gateway that lets AI agents create their own tools on the fly, so they can extend their capabilities instead of relying only on pre-built integrations.
Why it matters: Makes wafer-scale chips easier to program for scientific workloads.
Tungsten is a high-level language and compiler for Cerebras's Wafer-Scale Engine, released with Aviary, a set of scientific apps and tutorials. A coding agent wrote a plasma simulation that ran across 98% of a wafer.
Mapbox expanded its Meta partnership to power hands-free cycling navigation on Ray-Ban Display glasses, with bicycle-specific routing, spoken instructions and visual directions shown on the wearable display.
Rockwell Automation added Nvidia Omniverse, RTX rendering and OpenUSD to Emulate3D, its industrial simulation software, so manufacturers can model and test larger, more complex factory environments digitally.
RIG-TARS is a 3D-printable desktop robot from LuwuDynamics built on an ESP32-S3. Its three body segments, moved by SCS009 serial servos over UART, walk and turn on voice commands, and the design is open source.
The Xena Z2244 UALink traffic generator, available to buy, generates wire-speed traffic up to 800 Gb/s per device and injects errors to test UALink open interconnect gear in AI and HPC scale-up networks.
Richtech Robotics makes AI-driven service robots. It will show its new Intelligent Cleaning System at ISSA Show North America 2026, November 17-19 at Mandalay Bay in Las Vegas.
TOZO announced AIVU, AI glasses that bring intelligent assistance into everyday eyewear.
Why it matters: OS-level guardrails for agents set a precedent for how much AI apps can access.
macOS 27.3 beta implements tighter controls around Full Disk Access, letting users limit what AI automation apps can see and do on the Mac, following Apple's warning about increasingly capable autonomous AI agents.
Why it matters: A 457MW AMD-backed cloud deal expands OpenAI's compute beyond Nvidia and tests AMD as a frontier supplier.
TensorWave entered a six-year deal to provide OpenAI with AMD-powered cloud compute, adding 457MW of AI capacity as the two sides push back against an AI capex selloff.
Why it matters: Couples quantum hardware to GPU supercomputing scheduling, a step toward practical quantum.
Nvidia announced a collaboration with IBM and Fermilab on a quantum-classical codesign platform plus an error-correction breakthrough, while IQM and others reported work making quantum computers schedulable in Slurm via Nvidia's stack.
Why it matters: Adds sovereign European AI compute capacity as datacenter demand surges.
Nokia, Nscale, Skeleton, Telia and Verda formed AI Gigafactory Northern, planning a federated multi-site AI infrastructure network across the Nordics and Baltics and bidding to build it under a European tender.
A robotics company revealed a humanoid robot designed as a bodyguard and shared video of its capabilities.
Why it matters: Pushes coding agents and liquid-cooled infrastructure deeper into enterprise AI factories.
At SC26 Dell detailed direct liquid-cooled storage, Dell AI Data Platform advances and an OpenAI Codex integration for its AI Factory. A separate report says OpenAI added a remote switch for how Codex models run agent teams.
Why it matters: Extends Nvidia's roadmap from LLM training into agentic scientific computing.
Nvidia announced Rubin HPC, a new GPU to accelerate agents for science, engineering and analytics, alongside CUDA-X libraries giving agents scientific tools, Slurm expansion for quantum computing in supercomputing, and GPU-accelerated quantum error correction decoding.
A Hackster.io project, SafeRange, is a wearable that uses wireless channel sounding plus on-device AI to sense its surroundings, useful for proximity and safety alerts.
Vertiv's ATLAS Lab, announced November 16, 2026, tests AI factory power, cooling, controls and IT systems as one installation using emulated Nvidia computing, aiming to catch integration problems before production hardware arrives.
Nordic Semiconductor's nRF71 series are system-on-chip parts combining Wi-Fi 6, Bluetooth LE, Thread and on-device AI, with standby power up to 20 times lower than the current nRF70 line.
Why it matters: Moves co-packaged optics from demo to production in large AI training clusters.
Cloud provider Lambda is connecting Nvidia GB300 NVL72 racks with Nvidia Photonics CPO switches, which it calls the first large-scale production co-packaged optics deployment, improving networking efficiency and resiliency.
SwitchBot Hub 4 is a smart-home hub with a color display, physical controls, Hello Kata voice commands and Matter bridging for SwitchBot and infrared appliances.
Why it matters: Puts a purpose-built robotaxi service on a major ride-hailing network in a key US market.
Uber made Amazon's Zoox robotaxis available on its platform in Las Vegas, stepping up competition with Waymo, which also recently began operating there.
Why it matters: Extends Nvidia's physical-AI stack into edge reasoning for self-driving cars.
Nvidia's Alpamayo 2 Edge applies AI reasoning models to autonomous vehicles at the edge, aiming to improve how cars interpret and act on driving situations.
Why it matters: A 4.4 GW build is one of the largest announced AI data center commitments in India.
CtrlS outlines a 4.4 GW data center expansion in India as Seoul stocks open lower on AI concerns ahead of Nvidia earnings and a closely watched Anthropic IPO.
Why it matters: Shows how much forward capacity Nvidia has locked up, a signal for the whole AI buildout.
Tom's Hardware reports Nvidia amassed $530B in financial guarantees, with supply commitments climbing about a quarter of a trillion dollars in a single quarter to corner the AI market.
LG Energy Solution said it will attend RE+ 2026 in Las Vegas on Nov. 17-19, themed around energy for the AI era, and unveil an AI energy platform.
ROBOTERA's XHAND 1 Pro-U is a dexterous robotic hand now open for orders, featured in Nvidia's GR00T 2 DexUMI dexterous-manipulation demonstration.
In supported commercial AWS Regions, AgentCore harnesses created via API without explicit settings now use Claude Opus 5 for task instructions, context summarization and a managed built-in tool set AWS may revise.
Pydantic AI, a Python framework for building typed AI agents with realtime voice, image generation and embeddings across many models, ships v2.69.0. It folds RequestInput into HandleDeferredToolCalls, whose handler can take a resume token, and adds Gemini 4 Hydrogen support.
LLM is a command-line tool and Python library for running prompts against language models. Version 0.37.3 fixes a logging break caused by Homebrew shipping SQLite 3.54.1.
GoMirra is an AI companion service offering 18 characters with consent-based memory, multi-perspective conversations, voice chat and personalized daily briefings. It is launching publicly.
A KoreaTechDesk analysis describes how an enterprise AI assistant can trace a payment-failure incident to the wrong culprit when release notes and logs point at a recent deploy instead of an earlier promotion rule.
An open-source AI coding agent for VS Code that writes code from natural language, runs terminal commands and automates refactoring, with inline autocomplete and custom modes. It supports 500+ models including Claude and Gemini.
A merged SGLang fix made ten of ten Qwen tool calls schema-valid after nine of ten failed, and stopped GLM-4.5 from stringifying nullable arguments.
Genpact's Onyx is a platform for building, running and governing enterprise AI agents, using process intelligence data on workflows, policies and exceptions for context.
OpenAI is recruiting engineers to add travel booking and planning capabilities into ChatGPT, Skift reports.
Why it matters: Puts an agentic analyst inside a major BI platform.
Salesforce released Tableau Studio, an AI agent inside Tableau that helps users analyze data and build insights from their datasets.
SiliconANGLE highlights insights from GraphSummit 2026, including that AI agents need business context beyond enterprise data, and that knowledge graphs could unify that context into one layer for agents.
Google's Gemini Notebook now turns a user's notebook sources into shareable, interactive websites; Gemini skills also gain file support and sharing ahead of the Gems deprecation.
Perplexity's Monitor API gives developers scheduled topic searches and webpage-change tracking, with filtered updates via API or signed webhooks. It costs $1 per 1,000 successful runs.
A November 16 runbook in xAI's public workshop repo describes the cloud lifecycle for Cursor agents: working past a closed laptop, waiting on CI, and archiving the agent to end it.
A Reddit post describes miHoYo Gamer, a generalist agent built to play miHoYo games. No repository, benchmark or deployment details are given in the post.
League's AI Front Door for payers and AI Care Companion for providers are generally available on its Forge platform, pairing patient- and member-facing agents with safeguards and human escalation, deployable in as little as 12 weeks.
A merged vLLM patch compresses Qwen3.6's 16-bit lm_head from 970 MB to 300 MB at load time, cutting median decode time by 2.82 ms and gaining 12% on Strix Halo, with the perplexity cost published.
LangSmith is LangChain's platform for tracing and evaluating AI agents. The rebuild gives each registered agent a persistent, globally unique, immutable ID with a named owner, tying deployment, tracing, evaluation and governance together across runtimes.
Why it matters: Metering advanced Siri requests signals the cost structure Apple expects from on-device and cloud AI.
Apple's iOS 27.3, iPadOS 27.3 and macOS 27.3 betas introduce per-device daily caps on advanced Siri and Apple Intelligence requests while basic commands stay free. iCloud+ tiers raise limits; Safari gains an Ask About Page feature.
Grok Bot can now call Claude, though users cannot pick models, and Sabre announced a Grok Bot travel connector plus proactive agentic APIs; Grok Bot also appears on the Las Vegas Sphere.
ZDNET reports Samsung may bring a cross-app personal AI agent called Galaxy Claw to its phones, similar to Gemini Intelligence and Siri AI. Details on timing and availability are not yet confirmed.
Manus is an AI agent that automates workflows on your own computer. Version 2.1.5 is out for macOS, Windows and Linux.
AWS described a multi-stage method using Amazon Bedrock, Kiro IDE and AWS Transform to extract business rules from mainframe Assembler, convert via COBOL and deploy modernized Java services.
MinuteBox launched an AI agent that turns its entity management database into a 'system of work,' letting legal and compliance teams act on entity data.
Masttro, a wealth-data platform for family offices, released an MCP server and developer portal so AI assistants and agents can connect to its data, expanding AI interoperability for family offices.
Querit, a Singapore company founded in 2025, offers an agent retrieval API for AI search. It listed the API on AWS Marketplace so enterprise buyers can purchase it through existing AWS accounts and procurement.
At BoxWorks, Box introduced AI tools that help surface and retrieve information from unstructured content stores and build better process-automation workflows.
Indian AI firm Sarvam AI launched Token Factory, an inference service offering cheaper access to AI models served from India.
Amp is a command-line agent and development environment for coding tasks, installed via npm. This release is a routine build update with no documented features.
A Reddit user fed Claude their grandfather's 1945 WWII unit history book and had the model rebuild the platoon's war as an interactive 3D experience, making all 2,655 men in the book searchable.
tiny-voca is a 4.5M-parameter English text-to-speech model with two voices that runs about 70x faster than real time on four CPU threads; it ships alongside KLM-1.6, Nex-N2.6 and Edge0-145B-q3k releases on Hugging Face.
A dataset of Blender project files and videos behind 154 artworks by ARC, made 2019-2024, released under CC BY-NC 4.0 for non-commercial learning and remixing.
A dataset of 437 cleaned, redacted multi-step tool-use runs from OpenAI's Codex Desktop coding agent working in a local Windows workspace, plus 314 parent-to-subagent delegation examples.
Meituan open-sourced LongCat-2.5, a 1.6-trillion-parameter model for AI agents trained entirely on Chinese hardware, scoring 88.4 on Terminal Bench, just behind Claude Opus 5.
Pydantic AI, a Python framework for building typed AI agents with realtime voice, image generation and embeddings across many models, ships v2.69.0. It folds RequestInput into HandleDeferredToolCalls, whose handler can take a resume token, and adds Gemini 4 Hydrogen support.
turbo-browser-mcp is an open-source Playwright MCP server that lets AI agents control a browser with fewer turns and tokens, via one-call form filling, delta snapshots, batched steps and an on-by-default submit guard.
vLLM is an open-source high-throughput inference and serving engine for LLMs. Version 0.34.0 cuts a DeepSeek V4.1 Engram lookup from 54 ms to 5.1 ms on eight MI355X GPUs.
my-ai-coding-workflow is an agent-agnostic repo of AI coding instructions, curated skills and a three-stage review pipeline (planning, critical review, diff audit); it drew 203 stars in its first week.
copper-kicad-mcp is an MCP server that lets any AI app design real, orderable KiCad boards: schematic, LCSC parts, PCB routing, ERC/DRC checks and JLCPCB fab files. It is a new Python repo with 100 stars in its first week.
ViralForge is an Apache-2.0 skill, MCP server and CLI that recreates viral short videos, breaks hooks and scripts into a report, and ships ten costed variants.
Legwork is a Windows AI OSINT investigator: ask a question in plain words and get an answer built from public sources, with a citation on every sentence. The GitHub repo drew 301 stars in its first week.
GoMirra is an AI companion service offering 18 characters with consent-based memory, multi-perspective conversations, voice chat and personalized daily briefings. It is launching publicly.
OpenAI is recruiting engineers to add travel booking and planning capabilities into ChatGPT, Skift reports.
Google's Gemini Notebook now turns a user's notebook sources into shareable, interactive websites; Gemini skills also gain file support and sharing ahead of the Gems deprecation.
League's AI Front Door for payers and AI Care Companion for providers are generally available on its Forge platform, pairing patient- and member-facing agents with safeguards and human escalation, deployable in as little as 12 weeks.
Apple's iOS 27.3, iPadOS 27.3 and macOS 27.3 betas introduce per-device daily caps on advanced Siri and Apple Intelligence requests while basic commands stay free. iCloud+ tiers raise limits; Safari gains an Ask About Page feature.
Grok Bot can now call Claude, though users cannot pick models, and Sabre announced a Grok Bot travel connector plus proactive agentic APIs; Grok Bot also appears on the Las Vegas Sphere.
ZDNET reports Samsung may bring a cross-app personal AI agent called Galaxy Claw to its phones, similar to Gemini Intelligence and Siri AI. Details on timing and availability are not yet confirmed.
Manus is an AI agent that automates workflows on your own computer. Version 2.1.5 is out for macOS, Windows and Linux.
A Reddit user fed Claude their grandfather's 1945 WWII unit history book and had the model rebuild the platoon's war as an interactive 3D experience, making all 2,655 men in the book searchable.
WhatsApp Business now ships a built-in AI chatbot based on Meta's AI agent; Techzim walks through setting it up for business replies.
Luma Health released a ChatGPT plugin on November 16 that exposes 59 actions for health-system administrative work, letting the chatbot initiate clinic workflows instead of only answering questions.
Audible is introducing an AI-assisted discovery feature that combines audiobook samples, AI-generated overviews and author commentary so listeners can judge whether they will enjoy a title before starting it.
Durham University is giving every student access to Microsoft Copilot AI, Palatinate reports.
Tencent's Yuanbao AI assistant, which runs on Hunyuan and DeepSeek models, updates to version 2.93.0.605 on macOS, Windows and Linux.
Gamehaus says its internal AI pipeline reduces game prototyping work by about 90%, and its first AI-built game collection, Ease Games, is now live.
OpenAI is testing new ChatGPT search result formats including web results, See More and other layouts.
An XDA Developers blogger used Claude to create a tiny language model that boots from a floppy disk on a decades-old DOS CPU, plus Windows on an iPhone Duo.
Danilli released an AI app that lets users ask questions, shop, build websites and start a business from one assistant. It is available as an app.
Bank of America's 2026 Homebuyer Insights Report found 20% of prospective buyers use AI to shortlist homes, often before contacting an agent or visiting a lender's site.
Zoom Slides is an AI presentation maker that turns meeting transcripts, documents and ideas into editable decks with layouts, visuals and speaker notes, then exports to PowerPoint.
Cal.com released version 7 with Cal.com Events, a Discover directory, a Cal.ai Assistant for scheduling, and a new brand identity for the open scheduling platform.
Glifa is a Mac Markdown editor with AI inside the document: select text, press Shift-Cmd-R and ask for changes, with 18 built-in actions, live preview, math, Mermaid and Obsidian vaults; works with Claude Code, Codex, Apple Intelligence or a local model.
Blomma for Organizations is an AI career coach for managers and employees that knows company values, competencies and OKRs, helping with 1:1 prep, feedback, reviews and onboarding, and tracking goals privately. It launched on Product Hunt.
LinkedIn updated its comment detection and ranking systems to push AI-generated comments down in feeds. Early testing shows a 20%+ drop in views for such comments.
Huawei's Xiaoyi AI membership turns paid features into credits: 49 yuan for 1,000, Pro at 199 yuan for 6,000 and Max at 499 yuan for 15,000 a month.
In supported commercial AWS Regions, AgentCore harnesses created via API without explicit settings now use Claude Opus 5 for task instructions, context summarization and a managed built-in tool set AWS may revise.
LLM is a command-line tool and Python library for running prompts against language models. Version 0.37.3 fixes a logging break caused by Homebrew shipping SQLite 3.54.1.
A KoreaTechDesk analysis describes how an enterprise AI assistant can trace a payment-failure incident to the wrong culprit when release notes and logs point at a recent deploy instead of an earlier promotion rule.
An open-source AI coding agent for VS Code that writes code from natural language, runs terminal commands and automates refactoring, with inline autocomplete and custom modes. It supports 500+ models including Claude and Gemini.
A merged SGLang fix made ten of ten Qwen tool calls schema-valid after nine of ten failed, and stopped GLM-4.5 from stringifying nullable arguments.
Genpact's Onyx is a platform for building, running and governing enterprise AI agents, using process intelligence data on workflows, policies and exceptions for context.
Salesforce released Tableau Studio, an AI agent inside Tableau that helps users analyze data and build insights from their datasets.
SiliconANGLE highlights insights from GraphSummit 2026, including that AI agents need business context beyond enterprise data, and that knowledge graphs could unify that context into one layer for agents.
Perplexity's Monitor API gives developers scheduled topic searches and webpage-change tracking, with filtered updates via API or signed webhooks. It costs $1 per 1,000 successful runs.
A November 16 runbook in xAI's public workshop repo describes the cloud lifecycle for Cursor agents: working past a closed laptop, waiting on CI, and archiving the agent to end it.
A Reddit post describes miHoYo Gamer, a generalist agent built to play miHoYo games. No repository, benchmark or deployment details are given in the post.
A merged vLLM patch compresses Qwen3.6's 16-bit lm_head from 970 MB to 300 MB at load time, cutting median decode time by 2.82 ms and gaining 12% on Strix Halo, with the perplexity cost published.
LangSmith is LangChain's platform for tracing and evaluating AI agents. The rebuild gives each registered agent a persistent, globally unique, immutable ID with a named owner, tying deployment, tracing, evaluation and governance together across runtimes.
AWS described a multi-stage method using Amazon Bedrock, Kiro IDE and AWS Transform to extract business rules from mainframe Assembler, convert via COBOL and deploy modernized Java services.
MinuteBox launched an AI agent that turns its entity management database into a 'system of work,' letting legal and compliance teams act on entity data.
Masttro, a wealth-data platform for family offices, released an MCP server and developer portal so AI assistants and agents can connect to its data, expanding AI interoperability for family offices.
Querit, a Singapore company founded in 2025, offers an agent retrieval API for AI search. It listed the API on AWS Marketplace so enterprise buyers can purchase it through existing AWS accounts and procurement.
At BoxWorks, Box introduced AI tools that help surface and retrieve information from unstructured content stores and build better process-automation workflows.
Indian AI firm Sarvam AI launched Token Factory, an inference service offering cheaper access to AI models served from India.
Amp is a command-line agent and development environment for coding tasks, installed via npm. This release is a routine build update with no documented features.
Cohere's Video Intelligence in Compass lets AI agents retrieve specific timestamps across clips, transcripts and visual descriptions alongside documents; Cohere reports a 4.4-point lead over the strongest alternative in its video-ranking evaluation.
Nvidia's TensorRT-LLM more than doubled DeepSeek cache capacity and cut DeepSeek V4 Pro tuning time by 70% in a test, profiling 838 tactics instead of 82,160; the merged policy feature stays off by default. It also fixed a B300 race that sank R1 accuracy.
Virtana launched agentic observability that surfaces infrastructure insights inside Microsoft Teams, Salesforce Slack, and Anthropic's Claude.
Coral is an open-source gateway that lets AI agents create their own tools on the fly, so they can extend their capabilities instead of relying only on pre-built integrations.
Version 2.1.326 of Anthropic's Claude Code terminal assistant is out for the CLI and VS Code extension, letting Claude read codebases, edit files and run commands. A user report describes it leaking private file links during a refused security check.