A group of LEGO pros are hoping to build a less stressful environment for young patients by creating brick miniatures of MRI scanners, Xray machines and waiting rooms. Read...

The former West Coast player has described a culture and environment of “hypermasculinity” within the AFL that made questioning his sexuality impossible. Read...

TechHive Editors Choice At a glanceExpert`s Rating Pros Looks like an ordinary deadbolt   Can be incorporated into any smart home ecosystem   Wide variety of lock/unlock methods, including touch, fobs, smartphone, or smart watch   Cons Keypad is an added-cost option   Requires Level Connect or third-party Thread-capable bridge for Wi-Fi access   No biometric features High price tag   Our Verdict The Level Lock Pro is the best and most attractive smart lock you can buy, and its support for Matter means it can be seamlessly integrated into any smart home ecosystem.   Price When Reviewed This value will show the geolocated pricing text for product undefined Best Pricing Today Best Prices Today: Level Lock Pro Retailer Price Check Price comparison from over 24,000 stores worldwide Product Price Price comparison from Backmarket Best Prices Today: Check today’s prices We’re longtime fans of Level Lock’s family of smart deadbolts, the only products in their class that can pass for ordinary deadlocks, hiding all their intelligence in the motor that fits in a door’s bore hole. But the Level Lock Pro kicks things up a notch—maybe even a notch and half.   That said, the Matter smart home standard that’s designed to unify the fractured smart home space deserves a chunk of the credit. The Level Lock Touch, for instance, was the first Level Lock to allow you to lock and unlock the deadbolt with a touch of your hand, but Android users couldn’t operate the lock remotely over Wi-Fi.   Level eventually added support for Amazon’s Sidewalk, which required users to have an Echo smart speaker. Still later, Level introduced the Level Connect Wi-Fi bridge (2.4GHz networks only). Using either of those products finally put Android users on the same level as iPhone users (more precisely, iPhone users who also own a HomeKit hub such as a 2nd-gen Apple TV 4K or a HomePod smart speaker). And those aren’t your only options. There was a time when recommending a Level Lock required caveats. Thanks to Matter, that notion is firmly in the past.   Both Android and iPhone users will still need a bridge to connect the Level Lock Pro to their Wi-Fi networks to gain remote access and management, but both camps have a lot more options because the lock has a Thread radio onboard. Those options now include the previously mentioned Level Connect ($79), a Samsung SmartThings Station, an Apple TV 4K or HomePod, or any of the several Amazon Echo or Nest Hub smart speakers and displays with a Thread radio.   What’s even more important is that thanks to Matter, you no longer need to limit yourself to one smart home ecosystem or the other. An iPhone user like me can use an Apple TV or the like with the Level Lock Pro to take advantage of Apple’s Home Keys and also use a Thread-equipped Echo device or the Level Connect to control the lock with Alexa voice commands. Level tells me users can operate the new lock in as many as five smart home platforms simultaneously. While the Level Lock Pro doesn’t support Amazon’s Sidewalk, that really doesn’t matter.  Level Lock Pro: Specifications   Level Locks, including the unboxed Level Lock Pro shown here, have fewer components than other smart locks. Michael Brown/Foundry OK, enough preamble. Let’s turn our attention to the Level Lock Pro’s specifications. The new lock doesn’t look appreciably different from any of the earlier generations, although its exterior element is slightly larger in diameter. As usual, the lock’s motor fits inside the bore hole and is powered by a single CR2 lithium battery that’s cleverly hidden inside the lock’s bolt. The app reports remaining battery life as “Healthy,” “Low,” or “Critically Low,” at which point the battery won’t have enough power to lock or unlock the lock. Unlike some other smart locks, none of the locks in the Level series have a provision—such as a USB-C port—for providing temporary power to the lock to open it should its battery fail. That makes the inclusion of backup access in the form of a physical key even more important.  The Level Lock Pro boasts a BHMA AAA rating, which means it meets that standard body’s highest certification for security, durability, and finish respectively. Read our guide if you’d like to know more about the BHMA ratings for residential locks as well as the ANSI ratings for commercial locks. The Level Lock Pro has a weatherization rating of IP54, which means it will keep out enough dust to prevent its failure and that it’s protected from water sprayed from any direction. That’s relatively modest protection, but most entry doors get at least somewhat shielded from the elements. Read our IP code guide if you want more details on that topic.  Level says the Level Lock Pro also has a faster and more powerful microprocessor and faster memory than previous generations. Another new feature unique to the Pro is a sensor that can detect if the door is open or closed. This depends on using Level’s magnetic strike plate (the element you mount to the inside of your door’s frame). If you do, the lock will not throw its bolt if the door is ajar, preventing damage to your door trim when the bolt slams into it. The app will inform you of this status, and you can override the sensor and still throw the bolt if you need to (using either the thumb turn or the app), such as when you want to change the battery.   The Level Lock Pro’s motor makes physical and electrical contact with its bolt inside the bore hole (shown here) a puzzle-piece-shaped element.Michael Brown/Foundry As I’ve already discussed, The Level Lock Pro supports Matter-over-Thread, and it also has a Bluetooth radio. This is used for initial setup, but it also supports the optional Level Keypad ($79), a numeric keypad that lets you assign PINs to unlock the door. The lock also comes with two NFC (near-field communication) key fobs that you can put on a keychain. You can buy additional fobs in packs of four for $20, or you can buy four Level key cards for the same price (you can also re-program any existing Level fobs or key cards you might have from a previous installation). Finally, the lock comes with two physical keys.   The lock and its accompanying strike plate are available in either a matte black or satin nickel finish. Level tells me the lock and strike plate are fabricated from diecast zinc and stainless steel, with a ceramic coating that’s similar to powder coating, which all sounds very durable. Nonetheless, I managed to scratch the finish off the matte black strike plate during my installation (see photo), leading me to think it was merely painted black (Level disavowed me of that notion in response to a fact-check query).   Level Home says the Level Lock Pro and its strike plate have hardened finishes, but I managed to scratch the latter during installation.Michael Brown/Foundry While it’s unlikely that anyone will be using a physical key to unlock the door with regularity, I would recommend inserting the key into the key cylinder with a degree of precision, lest the same thing happen to the lock, despite its ceramic coating.  Level Lock Pro: Installation and setup   As I’ve experienced with previous Level Locks, the Pro is supremely easy to install and set up, and Level’s four-part installation card explains everything well—with one exception. Out of the box, the interior paddle you’ll turn to retract the lock’s bolt is attached to its interior escutcheon, and it must be removed to expose the holes into which you’ll insert the screws that will fasten the lock’s interior element to its exterior.   Accomplishing this isn’t immediately obvious, and despite having installed and reviewed the Level Lock+ about this time two years ago, I’d forgotten how to do it. Only after scanning the user manual twice did I notice a footnote reading “To disassemble in the future, use a paperclip or pen to press the small hole in the bottom of the paddle, while gently pulling away from the door to release it.”   The Level Lock Pro motor inside the bore hole and secured to its bolt. The two are secured with a small screw on the opposite side.Michael Brown/Foundry Had it not been for that delay, I would have finished the physical installation in less than 10 minutes. You don’t need to engage in any delicate balancing act to hold the lock’s interior and exterior elements together, and you don’t need to thread a power cable from the lock’s exterior escutcheon through the door’s bore hole and then plug it into a socket on the interior escutcheon, as is common with so many other smart locks.   You’ll want to install the Level app on your phone, add the lock to the app, and download and install any firmware updates before you scan the QR code to onboard it to your Matter-certified smart home hub; otherwise, you’ll have only a 30-minute window to add the lock to the Level app. Miss that window and you’ll need to hardware reset the lock and start over. This isn’t a flaw, it’s just a matter of not putting the cart in front of the horse, and it’s clearly spelled out in the installation manual.   You’ll need the Level app to accomplish some tasks, such as creating PINs, setting user permission levels, programming fobs and key cards, and pairing accessories such as the keypad. Other jobs, such as creating Apple home keys, will be performed in their respective apps—Apple Home in that example.   Level Lock Pro: Using the Level Lock Pro   I reviewed the Level Lock Pro using both the Level Connect and a 2nd-gen Apple TV 4K.Michael Brown/Foundry I tested the Level Lock Pro with the Level Connect bridge first and then added it to my Apple Home environment, connecting it to a 2nd-gen Apple TV 4K. Apart from the new door-sensing feature, the Level Lock Pro operates just like previous models, if perhaps a bit faster to respond thanks to its new processor. When you exit, one touch on the lock’s exterior body throws the bolt and sounds a chime to let you know the door is locked (a chime with a different tone sounds when the door is unlocked).   When you return, you can use a key fob or card, the Level app on your iPhone or Android smartphone, the NFC features on your smartphone or smart watch, or a physical key to unlock it. An auto-lock feature automatically locks the door after it’s been unlocked, and you can choose the amount of time that passes before that happens, from 30 seconds to one hour. You can also choose to disable this feature.  The Matter-certified Level Connect bridge plugs into any electrical outlet.Michael Brown/Foundry The Level Lock Pro supports geofencing, and you can enable a setting that allows you to unlock the lock by touching it when you return home. This option is timed—your choice of one, two, or three minutes—and you won’t be able to use the touch method after that time expires. Also note that you must first cross over the border of your geofence before a touch-to-unlock can be performed; it won’t work if you just go out into your yard and then come back after the lock is locked.   The Level app logs the date and time when the lock was locked, unlocked, and the door is opened; how the action was performed (manually, via touch, key fob, or keypad); as well as who unlocked the door (key fobs and cards can be assigned to individuals, although the lock obviously can’t know who actually used the device to unlock the door).   The optional keypad, available in colors to match your lock, has a capacitive touch surface with LED backlighting (these aren’t cheap rubber buttons). It runs on two AAA batteries (included) that Level says should last up to two years. You can create up to 50 unique PINs and assign them limited-time permissions in the Level app if you so desire. You can also lock a paired Level Lock with this accessory, although you’ll find yourself more typically doing this by touching the lock itself if you’ve enabled that feature.  The Level keypad features capacitive touch buttons that light up when touched.Michael Brown/Foundry The keypad needs to be installed within 30 feet of the lock, which shouldn’t present any obstacles. It’s thoroughly weatherized—IP65—which according to our IP code guide means it’s impervious to dust ingress and that it can withstand water jets projected from any direction (short of a pressure washer, but you should be able to hose it down to clean it if needed). It has an operating temperature range of negative four degrees to 140 F (-20 to +60 C).  Level Lock Pro: User management   You can grant permission to other people to use the lock as an Admin (who can control every aspect of the lock, including inviting new users) or as a Guest (with three categories of limited control). When you set up their access, the Level app will send an invitation to their mobile phone and/or email.  The three levels of Guest access are Always, Temporary, or Recurrent. You’d probably want to give family members the Guest/Always level of permission, which lets them unlock the door at any time of day or night.   An exploded view of the Level Lock Pro.Level Home With Guest/Temporary access, you set start and end dates and times, which would be useful for vacation rentals (e.g., from 11:00 a.m. on August 29, 2025 to 3:00 p.m. on August 29). Recurrent access is intended for people to whom you wish to grant regular access to your home, but only at set times on given days of the week. If you have a housecleaning service that comes to your house every Thursday at 10:00 a.m., for instance, they’d be able to unlock the door.   If you also have the Level keypad, you can create PINs for each person to whom you want to grant access to your home. These can be up to six digits long, but PINs have only two levels of access: Always or Temporary (i.e., the PIN will only work during a range of dates and times). Those permissions can be revoked at any time.  Level Lock Pro: Conclusion There was a time when recommending a Level Lock required several caveats, the most significant being that Android users were treated as second-class users denied access to some of its best features. Thanks to Matter support and the Thread radio in this lock (and in the earlier Level Lock+) that notion is firmly in the past.   As with previous generations of this smart lock, the Level Lock Pro is the most attractive smart lock you can buy, primarily because it doesn’t look like a smart lock. That said, this is a pricey smart lock at $349, and you’ll need to pay extra if you want to use a numeric keypad to unlock the door, a feature that comes standard on many of Level’s competitors. The fact that you’ll need either a Level Connect or a third-party device with a Thread radio to gain out-of-home control adds another layer of expense—unless you already own such a device. Level also doesn’t provide any means of biometric access–a fingerprint reader or a palm-vein scanner–as a number of its competitors now offer (albeit in much larger form factors). Even with all those considerations, the Level Lock Pro is still the best smart lock on the market. No competing product is prettier or easier to install and use on a day-to-day basis. I recommend it without reservation.    This review is part of TechHive’s in-depth coverage of the best smart locks. Read...

Arlo announced a refresh of its entire home security camera lineup today, unveiling new models in its Essential (HD and 2K resolution), Pro (2K resolution), and Ultra (4K) series. The company also revealed its very first pan/tilt models that can spin 360 degrees and tilt 180 degrees to cover every angle. The company says each of the new cameras features Arlo Intelligence, which will help Arlo users get a better understanding of what’s happening in and around their homes via personalized, actionable alerts. New pan/tilt cameras The biggest news is Arlo’s long-overdue first pan/tilt cameras, the Essential Pan Tilt Indoor Camera and the weather-resistant Essential Pan Tilt Security Camera. Both models pan a complete 360 degrees and tilt 180, ensuring full coverage of an entire room or yard. They can automatically track motion, switch between 2.4- and 5GHz Wi-Fi bands to achieve the strongest network connection. Other standard features include two-way audio and a built-in siren; the outdoor model adds a spotlight for color night vision, which can help to capture details such as license plates in low light. Privacy controls in the Arlo Secure app can disable audio and video when you don’t want eyes or ears on. The Essential Pan Tilt Indoor Camera is priced at $34.99 for the HD (1080p) model and $49.99 for the 2K version. The outdoor Essential Pan Tilt Security Camera lists for $49.99 with HD resolution and $59.99 with 2K res. The Arlo Essential Pan Tilt Indoor Camera pans 360-degrees and tilts 180, ensuring full coverage of the environment.Arlo The Arlo Essential series gets a refresh Arlo has also refreshed its entry-level Essential line. The third-generation models also come in both HD and 2K versions, with options for plug-in or battery power. They cover the basics: indoor monitoring, outdoor security, or extended runtime with the larger batteries in the XL models. Here’s how the lineup breaks down: Essential Indoor Camera (3rd Gen) Plug-in: $24.99 (HD) or $39.99 (2K) Essential Security Camera Plug-in: $39.99 (HD) or $49.99 (2K) Essential Security Camera (3rd Gen) Battery Operated: $49.99 (HD) or $69.99 (2K) Essential XL Security Camera (3rd Gen) Battery Operated: $69.99 (HD) or $79.99 (2K) Arlo Pro series advances to its 6th generation The 6th-generation Arlo Pro series sits in the middle of the new lineup. These cameras record in 2K HDR, deliver a wider view (160 degrees), and add color night vision for clearer detail after dark. They also use swappable rechargeable batteries, so you can keep one charging while the other is inside the camera. The Pro model is priced at $124.99 and the XL version–which has a larger battery–costs $199.99. The Ultra series tops the range At the top end is the third-generation Ultra. It records in 4K HDR with a 180-degree field of view, pairing sharp detail with wide coverage. The camera is built to handle heat, cold, rain, and snow, and it includes two-way audio with noise reduction for clearer conversations. It lists for $199.99. Built to handle the elements, the Essential Pan Tilt Security Camera adds a spotlight that reveals details like clothing, license plates, and faces in low light. Arlo Arlo’s AI at the center Each of the new cameras works with Arlo Secure, the subscription service that uses AI to flag people, vehicles, packages, and hazard warnings from devices such as smoke alarms. It can also add captions to events to provide quick context. Plans start at $7.99 a month, with a premium option that includes 24/7 monitoring and one-tap access to emergency responders. Each of the new Arlo cameras is available for pre-order now through Arlo’s website and Amazon. We’ll have reviews of the new family soon, starting with the two pan-tilt models. Read...

Environment Southland chair Nicol Horrell has defended speeding in his work car, saying it was mostly for overtaking. Read...

The advice was mistakenly released by the Ministry for the Environment. Read...

How badly does AI harm the environment? We now have some answers to that question, as both Google and Mistral have published their own self-assessments of the environmental impact of an AI query. In July, Mistral, which publishes its own AI models, published a self-evaluation of the environmental impact of training and querying its model in terms of the amount of carbon dioxide (CO2) produced, the amount of water consumed, and the amount of material consumed. Google took a slightly different approach, publishing the amount of power and water a Gemini query consumes, as well as how much CO2 it produces. Of course, there are caveats: Each report was self-generated, and not performed by an outside auditor. Also, training a model consumes vastly more resources than inferencing, or the day-to-day tasks users assign a chatbot each time they query it. Still, the reports provide some context for how much AI taxes the environment, even though they exclude the effects of AI training and inferencing by OpenAI and other competitors. On Thursday, Google said its estimate for the resources consumed by a “median” Gemini query consumes 0.24Wh of energy and 0.26 milliliters (five drops) of water, and generates the equivalent of 0.03 grams of carbon dioxide — the equivalent of 9 seconds of watching TV. Mistral’s report slightly differed: For a “Le Chat” response generating a page of text (400 tokens), Mistral consumes 50 milliliters of water, produces the equivalent of 1.14 grams of carbon dioxide, and consumes the equivalent of 0.2 milligrams of non-renewable resources. Google said “comparative models” typically are a bit more lenient, and only look at the impacts of active TPU and GPU consumption. Put this way, the median Gemini text prompt uses 0.10Wh of energy, consumes 0.12ml of water, and emits the equivalent of 0.02 grams of carbon dioxide. Google did not release any assessments of the impact of training its Gemini models. Mistral did: In January 2025, training its Large 2 model produced the equivalent of 20.4 kilotons of carbon dioxide, consumed 281,000 cubic meters of water, and consumed 650 kilograms of resources. That’s about 112 Olympic-sized swimming pools of water consumption. Using the EPA’s estimate that an average car produces 4.6 metric tons of carbon dioxide annually, that works out to the annual CO2 production of 4,435 cars, too. The environmental impact assessments assume that energy is produced via means that actually produce carbon dioxide, such as coal. “Clean” energy, like solar, lowers that value. Likewise, the amount of water “consumed” typically assumes the use of evaporative cooling, where heat is transferred from the chip or server (possibly being cooled by water as well) to what’s known as an evaporative cooler. The evaporative cooler transfers heat efficiently, in the same manner as your body cools itself after a workout. As you sweat, the moisture evaporates, an endothermic reaction that pulls heat from your body. An evaporative cooler performs the same function, wicking heat from a server farm but also evaporating that water back into the atmosphere. Mistral’s environmental impact assessment includes a footnote noting the differences in electricity France and the United States consume. Google said that it uses a holistic approach toward managing energy, such as more efficient models, optimized inferencing though models like Flash-Lite, custom-built TPUs, efficient data centers, and efficient idling of CPUs that aren’t being used. Clean energy generation — such as a planned nuclear reactor — can help lower the impact numbers, too. “Today, as AI becomes increasingly integrated into every layer of our economy, it is crucial for developers, policymakers, enterprises, governments, and citizens to better understand the environmental footprint of this transformative technology,” Mistral’s own report adds. “At Mistral AI, we believe that we share a collective responsibility with each actor of the value chain to address and mitigate the environmental impacts of our innovations.” How much water and electricity does ChatGPT consume? The reports from Mistral and Google haven’t been duplicated by other companies. EpochAI estimates that the average GPT-4o query on ChatGPT consumes about 0.3Wh of energy, based upon its estimates of the types of servers OpenAI uses. However, the amount of resources AI consumes can vary considerably, and even AI energy scores are rudimentary at best. “In reality, the type and size of the model, the type of output you’re generating, and countless variables beyond your control—like which energy grid is connected to the data center your request is sent to and what time of day it’s processed—can make one query thousands of times more energy-intensive and emissions-producing than another,” an MIT Technology Review study found. Its estimates of 15 queries a day plus 10 images plus three 5-second videos would consume 2.9kWh of electricity, it found. Still, Mistral’s study authors note that its own estimates point the way toward a “scoring system” where buyers and users could use these studies as a way to choose AI models with the least environmental impact. It also called upon other AI model makers to follow its lead. Whether AI is “bad” for the environment is still up for discussion, but the reports from Google and Mistral provide a foundation for a more reasoned discussion. Read...

NZ facing its “most challenging security environment” in years from foreign interference, espionage, and more young people being radicalised online. Read...

Foreign interference, espionage, and online radicalisation threats have increased since last year, the Security Intelligence Service says. Read...

Modern notebooks with integrated AI hardware are changing the way artificial intelligence is used in everyday life. Instead of relying on external server farms, these large language models, image generators, or transcription systems run directly on the user’s own device. This is made possible by the combination of powerful CPUs, dedicated graphics processors and, at the center of this development, a Neural Processing Unit (NPU). An NPU is not just an add-on, but a specialized accelerator designed precisely for the calculation of neural networks. It enables offline AI tools such as GPT4All or Stable Diffusion not only to start, but also to react with high performance, low energy consumption and constant response time. Even with complex queries or multimodal tasks, the working speed remains stable. The days when AI was only conceivable as a cloud service are now over. Work where others are offline As soon as the internet connection is interrupted, classic laptops begin to idle. An AI PC, on the other hand, remains operational, whether in airplane mode above the clouds, deep in the dead zones of rural regions, or in an overloaded train network without a stable network. In such situations, the structural advantage of locally running AI systems becomes apparent. Jan.ai or GPT4All can be used to create, check and revise texts, intelligently summarize notes, pre-formulate emails and categorize appointments. Foundry With AnythingLLM, contracts or meeting minutes can be searched for keywords without the documents leaving the device. Creative tasks such as creating illustrations via Stable Diffusion or post-processing images with Photo AI also work, even on devices without a permanent network connection. Even demanding projects such as programming small tools or the automated generation of shell scripts are possible if the corresponding models are installed. For frequent travelers, project managers, or creative professionals, this creates a comprehensive option for productive working, completely independent of infrastructure, network availability, or cloud access. An offline AI notebook does not replace a studio, but it does prevent downtime. Sensitive content remains local Data sovereignty is increasingly becoming a decisive factor in personal and professional lives. Anyone who processes business reports, develops project ideas, or analyzes medical issues cannot afford to have any uncertainties when processing data. Public chatbots such as Gemini, ChatGPT, or Microsoft Copilot are helpful, but are not designed to protect sensitive data from misuse or unwanted analysis. Local AI solutions, on the other hand, work without transmitting data to the internet. The models used, such as LLaMA, Mistral or DeepSeek, can be executed directly on the device without the content leaving the hardware. This opens up completely new fields of application, particularly in areas with regulatory requirements, such as healthcare, in a legal context, or in research. AnythingLLM goes one step further. It combines classic chat interaction with a local knowledge base of Office documents, PDFs and structured data. This turns voice AI into an interactive analysis tool for complex amounts of information, locally, offline and in compliance with data protection regulations. NPU notebooks: new architecture, new possibilities While traditional notebooks quickly reach their thermal or energy limits in AI applications, the new generation of copilot PCs rely on specialized AI hardware. Models such as the Surface Laptop 6 or the Surface Pro 10 integrate a dedicated NPU directly into the Intel Core Ultra SoC, supplemented by high-performance CPU cores and integrated graphics. The advantages are evident in typical everyday scenarios. Voice input via Copilot, Gemini or ChatGPT can be analyzed without delay, image processing with AI tools takes place without cloud rendering, and even multimodal tasks, such as analyzing text, sound, and video simultaneously run in real time. Microsoft couples the hardware closely with the operating system. IDG Windows 11 offers native NPU support, for example for Windows Studio Effects, live subtitles, automatic text recognition in images or voice focus in video conferences. The systems are designed so that AI does not function as an add-on, but is an integral part of the overall system as soon as it is switched on, even without an internet connection. Productive despite dead zones The tools for offline AI are now fully developed and stable in everyday use. GPT4All from Nomic AI is particularly suitable for beginners, with a user-friendly interface, uncomplicated model management and support for numerous LLMs. Ollama is aimed at technically experienced users and offers terminal-based model management with a local API connection, ideal for providing your own applications or workflows directly with AI support. LM Studio, on the other hand, is characterized by its GUI focus. Models from Hugging Face can be simply be searched in the app, downloaded, and activated with a click. The LM Studio chatbot not only provides access to a large selection of AI models from Huggingface.com, but also allows the AI models to be fine-tuned. There is a separate developer view for this. LM Studio Jan.ai is particularly versatile. The minimalist interface hides a highly functional architecture with support for multiple models, context-sensitive responses, and elegant interaction. Local tools are also available in the creative area. With suitable hardware, Stable Diffusion delivers AI-generated images within a few seconds, while applications such as Photo AI automatically improve the quality of screenshots or video frames. A powerful NPU PC turns the mobile device into an autonomous creative studio, even without Wi-Fi, cloud access, or GPU calculation on third-party servers. What counts on the move The decisive factor for mobile use is not just whether a notebook can run AI, but how confidently it can do this offline. In addition to the CPU and GPU, the NPU plays a central role. It processes AI tasks in real time, while at the same time conserving battery power and reducing the load on the overall system. Devices such as the Galaxy Book with an RTX 4050/4070 or the Surface Pro 10 with a Intel Core Ultra 7 CPU demonstrate that even complex language models such as Phi-2, Mistral, or Qwen run locally, with smooth operation and without the typical latencies of cloud services. Copilot as a system assistant complements this setup, provided the software can access it. When travelling, you can compose emails, structure projects, prepare images or generate text modules, regardless of the network. Offline AI on NPU notebooks also transforms the in-flight restaurant, the waiting gate, or the remote holiday home into a productive workspace. Requirements and limitations The hardware requirements are not trivial however. Models such as LLaMA2 or Mistral require several gigabytes of RAM, 16 GB RAM is the lower minimum. Those working with larger prompts or context windows should plan for 32 or 64 GB. The SSD memory requirement also increases, as many models use between 4 and 20 GB. NPUs take care of inference, but depending on the tool, additional GPU support may be necessary, for example for image generation with Stable Diffusion. Sam Singleton Integration into the operating system is also important. Copilot PCs ensure deep integration between hardware, AI libraries, and system functions. Anyone working with older hardware will have to accept limitations. The model quality also varies. Local LLMs do not yet consistently reach the level of GPT-4, but they are more controllable, more readily available and more data protection-friendly. They are the more robust solution for many applications, especially when travelling. Offline AI under Linux: openness meets control Offline AI also unfolds its potential on Linux systems—often with even greater flexibility. Tools such as Ollama, GPT4All, or LM Studio offer native support for Ubuntu, Fedora, and Arch-based distributions and can be installed directly from the terminal or as a flatpack. The integration of open models such as Mistral, DeepSeek, or LLaMA works smoothly, as many projects rely on open source frameworks such as GGML or llama.cpp. Browser interface for Ollama: Open-Web-UI is quickly set up as a Python program or in a Docker container and provides a user interface.IDG Anyone working with Docker or Conda environments can build customized model set-ups, activate GPU support or fine-tune inference parameters. This opens up various scenarios, especially in the developer environment: Scripting, data analysis, code completion, or testing your own prompt structures. In conjunction with tiling desktops, reduced background processes and optimized energy management, the Linux notebook becomes a self-sufficient AI platform, without any vendor lock-in, with maximum control over every file and every computing operation. Offline instead of delivered Offline AI on NPU notebooks is not a stopgap measure, but a paradigm shift. It offers independence, data protection, and responsiveness, even in environments without a network. Thanks to specialized chips, optimized software, and well thought-out integration in Windows 11 and the latest Linux kernel, new freedom is created for data-secure analyses, mobile creative processes, or productive work beyond the cloud. The prerequisite for this is an AI PC that not only provides the necessary performance, but also anchors AI at a system level. Anyone relying on reliable intelligence on the move should no longer hope for the cloud, but choose a notebook that makes it superfluous. Read...