Microsoft is rolling out a major Notepad update that finally brings true on-device generative AI to Windows 11 — allowing users of Copilot+ PCs to generate, rewrite, and summarize text locally with no Microsoft 365 subscription or constant internet connection required.
Notepad’s AI capabilities have evolved quickly from cloud‑only helpers into a hybrid system that can run local language models on compatible hardware. The new update adds three writing-assistance modes — Write, Rewrite, and Summarize — that can operate using on‑device models when the PC qualifies as a Copilot+ PC. That removes the need to be signed into a Microsoft account or to consume cloud-based AI credits for basic generation and rewriting tasks, while still letting users switch to cloud models when they want more power or up‑to‑date knowledge.
This change is notable because it moves a core Windows inbox app from a cloud-dependent helper into a hybrid local-first assistant, shifting the conversation about AI on Windows from always online to capability-based and privacy‑aware for supported devices.
Microsoft’s strategy for delivering local AI hinges on the Copilot+ PC concept: machines equipped with a Neural Processing Unit (NPU) capable of high throughput (40+ TOPS). These NPUs are designed to run optimized language and multimodal models locally, enabling not only responsiveness and offline operation but also allowing Microsoft to offer a subset of generative functionality without subscription gating.
The most recent rollout puts local Notepad AI into preview for Windows Insiders in the Canary and Dev channels, with broader availability expected to follow in the near term for qualifying devices.
NPUs are optimized for low-precision matrix math and transformer inference patterns, which lets manufacturers deliver respectable local model performance in mobile laptops and fan-optimized designs. This hardware investment is what makes local models viable in consumer notebooks, moving these features from data center‑bound experiments into everyday apps.
There are real limitations — hardware exclusivity, model capability, and update/telemetry transparency are the most important. For users with Copilot+ hardware, the result is likely to be a fast, private, and useful set of writing helpers built into a familiar app. For everyone else, the update is a reminder that not all AI features are created equal: access will increasingly hinge on device class and vendor partnerships.
This is a welcome evolution for Notepad and for Windows as a platform: a move toward local intelligence that respects both utility and privacy while keeping the door open for cloud power when needed.
Source: Windows Central Windows 11 Notepad will soon let you generate text using on-device AI models — no subscription required
Overview
Notepad’s AI capabilities have evolved quickly from cloud‑only helpers into a hybrid system that can run local language models on compatible hardware. The new update adds three writing-assistance modes — Write, Rewrite, and Summarize — that can operate using on‑device models when the PC qualifies as a Copilot+ PC. That removes the need to be signed into a Microsoft account or to consume cloud-based AI credits for basic generation and rewriting tasks, while still letting users switch to cloud models when they want more power or up‑to‑date knowledge.This change is notable because it moves a core Windows inbox app from a cloud-dependent helper into a hybrid local-first assistant, shifting the conversation about AI on Windows from always online to capability-based and privacy‑aware for supported devices.
Background: How Notepad arrived at on‑device generation
Notepad — once the lean plain‑text editor bundled with Windows — has been modernized over the last year with features that were long overdue: tabs, basic formatting, markdown support, and now integrated Copilot-style writing tools. Initially, Notepad’s AI features (Summarize, Rewrite, Write) were cloud-driven. They required a Microsoft account and surfaced behind a credit/subscription model in some regions.Microsoft’s strategy for delivering local AI hinges on the Copilot+ PC concept: machines equipped with a Neural Processing Unit (NPU) capable of high throughput (40+ TOPS). These NPUs are designed to run optimized language and multimodal models locally, enabling not only responsiveness and offline operation but also allowing Microsoft to offer a subset of generative functionality without subscription gating.
The most recent rollout puts local Notepad AI into preview for Windows Insiders in the Canary and Dev channels, with broader availability expected to follow in the near term for qualifying devices.
What “on‑device” actually means for Notepad
Hardware requirements and the Copilot+ PC designation
Notepad’s local AI requires a Copilot+ PC — a class of Windows 11 machines that include an NPU capable of performing 40+ TOPS. Typical Copilot+ processors and platforms include modern Snapdragon X-series SoCs and recent waves of silicon branded for AI acceleration (examples span Intel Core Ultra series variants and AMD Ryzen AI lines on supported SKUs). Copilot+ hardware requirements usually include:- An NPU with 40+ TOPS capability
- A modern CPU and GPU combination compatible with Windows 11 Copilot+ features
- Memory and storage meeting Copilot+ class minimums (commonly 16 GB RAM and 256 GB SSD as baseline on many Copilot+ specs)
Software and model runtime
When the on‑device mode is selected, Notepad uses model runtimes that run locally on the NPU. The OS will manage downloading or provisioning the model runtime and model weights as needed (subject to device storage and admin controls). Users can switch between:- Local models — run entirely on the device NPU, usable offline, and available without a Microsoft account or subscription.
- Cloud models — run through Microsoft’s cloud services, offering potentially larger models, fresher knowledge, or features gated behind subscriptions and AI credits.
What users can do in Notepad with on‑device AI
Notepad’s AI toolkit provides three core capabilities tailored for common writing tasks:- Write: Generate new text from a prompt or expand a selected reference. Useful for drafting emails, notes, lists, or initial ideas.
- Rewrite: Rephrase selected text with options to change tone, length, or format. Notepad can return multiple variants so users choose the best fit.
- Summarize: Produce short, medium, or long summaries of highlighted content to distill documents into concise takeaways.
- Turn rough bullet points into a coherent paragraph for a quick email.
- Rephrase a paragraph to sound more formal or more concise for different audiences.
- Summarize meeting notes into a short action item list, even when offline.
How to use the new features (practical steps)
- Open Notepad on a Copilot+ PC with the updated app (Insider preview or later public build).
- To generate text (Write):
- Place the cursor where you want new content, or select reference text.
- Open the Copilot menu or right‑click and choose “Write.”
- Type your prompt, adjust settings, and insert the output.
- To rewrite text:
- Highlight the text to change.
- Use the toolbar Rewrite button, context menu, or the app’s Rewrite shortcut.
- Choose tone/length options and select one of multiple generated variants.
- To summarize:
- Select text and choose Summarize from the Copilot menu or context menu (or use the Summarize shortcut).
- Pick desired summary length and insert.
Security and privacy implications
On-device generation changes the privacy calculus in several meaningful ways:- Local data processing: When Notepad uses the local model, the text you generate or edit is processed on the device, not sent to Microsoft’s cloud for inference. That reduces the attack surface for content exfiltration in transit and provides a stronger privacy posture for sensitive drafts.
- Model provisioning and updates: While inference happens locally, model binaries and runtimes are typically delivered and updated by Microsoft (or the OEM). That process may involve downloads and telemetry; enterprises and privacy‑conscious users should check update policies and MDM controls.
- Administrative control: Organizations can control whether on‑device models are allowed, which models are provisioned, and how updates are applied through standard Windows management tooling. For regulated environments, local processing can be a compliance advantage — provided provisioning and update flows are configured properly.
- No subscription barrier for local use: Removing the subscription requirement for local inference lowers friction but also shifts trust: users must trust device-level protections and model behavior rather than cloud policy enforcement.
Performance, battery, and UX trade‑offs
Running language models locally on an NPU offers responsiveness and offline capability, but there are practical trade‑offs:- Latency and speed: Local inference on NPUs will usually be faster than cloud roundtrips for short tasks, and streaming output can feel instantaneous. However, model size and NPU thermal constraints may throttle throughput for long or repeated generations.
- Battery and thermals: Heavy on‑device inference increases power draw and can raise device temperatures, potentially affecting sustained performance and battery life. Copilot+ vendors optimize hardware and firmware for these loads, but users should expect heavier power usage during extended AI sessions.
- Model capability: On‑device models are typically smaller and optimized for NPU execution; they may not match the depth or recentness of cloud models. For highly specialized or up‑to‑the‑minute knowledge, the cloud option remains preferable.
- Storage: Model runtimes and weights consume several hundred megabytes to multiple gigabytes depending on model complexity. Devices with limited storage may prompt users to manage or remove models.
Business and enterprise considerations
The arrival of on‑device AI in a ubiquitous app like Notepad has nontrivial implications for IT departments:- Secure offline workflows: On‑device inference enables secure content generation without internet connectivity — valuable for air‑gapped or high‑security contexts.
- Patch and model lifecycle: Enterprises need policies for model provisioning, signing, and updates. Centralized controls can ensure only approved models are used and that updates are staged before broad deployment.
- User training and consent: Employees must be instructed about when local models are allowed and where cloud processing may still occur (for example, when a user explicitly selects cloud generation).
- Licensing and support: While local functionality lowers subscription friction for basic tasks, cloud features and expanded quotas may still require licensing. IT should map use cases to licensing models to avoid unexpected costs.
- Endpoint management: MDM suites and security stacks must be configured to monitor model installations, binaries, and runtime telemetry to retain visibility into on‑device AI behavior.
Strengths of Microsoft’s approach
- Local-first privacy: Allowing inference entirely on device significantly improves privacy for users and organizations who prefer to keep content off the cloud.
- No subscription barrier for basic use: Removing the Microsoft 365 sign‑in and subscription requirement for on‑device use lowers friction and makes AI writing assistance more widely accessible on compatible hardware.
- Hybrid flexibility: Letting users switch between local and cloud models provides practical flexibility: use local for quick, private edits; cloud for heavier, knowledge‑rich tasks.
- Integration into an everyday app: Notepad is familiar and lightweight; embedding AI here lowers the barrier to experimentation without forcing users into heavier, subscription‑locked tools.
Risks, limitations, and unknowns
- Hardware exclusivity: The Copilot+ PC requirement fragments access. Many recent “AI” laptops lack the exact NPU spec Microsoft demands, which will frustrate users who expected on‑device AI from their existing devices.
- Model fidelity: On‑device models are optimized for size and speed and may underperform against large cloud models in nuance, factuality, and reasoning depth.
- Telemetry and updates: Even with local inference, model provisioning and telemetry remain areas of concern. The exact telemetry surface for model downloads and health reporting is not always transparent in early previews.
- Supply chain and vendor lock‑in: Copilot+ branding ties particular AI experiences to a combination of Microsoft software and OEM hardware. That can complicate cross‑vendor parity and raise questions about long‑term availability on devices from certain vendors.
- Security of model binaries: Local models add a new attack surface: signed model files and runtimes. Ensuring model authenticity and secure provisioning is critical.
- Usability inconsistencies: Early Insider builds frequently vary in keyboard shortcuts, UI flows, and feature availability by region. Users should expect refinement over time.
How this changes the consumer AI landscape on Windows
Notepad’s move to support local generative models marks an important shift in how Microsoft positions AI across Windows:- It establishes a practical baseline for on‑device AI experiences in everyday utilities.
- It forces OEMs and silicon partners to standardize around stronger NPU capabilities if they want to deliver the full Copilot+ experience.
- It sets a precedent that Microsoft can provide some generative AI capabilities without subscription, while still preserving subscription value for heavier cloud features.
Practical recommendations for users
- If privacy and offline operation matter: Obtain a Copilot+ PC or confirm your device meets the NPU and hardware requirements to use local Notepad AI.
- If you want maximum model capability and freshness: Use cloud generation for tasks that rely on current events, web knowledge, or deep reasoning, and be prepared for potential subscription / credit usage for high-volume scenarios.
- For IT admins: Assess Copilot+ features against compliance needs, define model provisioning policies, and use MDM to control model installations and telemetry.
- To test now: Join the Windows Insider Program’s Canary or Dev channels if you want early access, but be prepared for instability and changing UI/shortcuts across builds.
The broader technical picture: why NPUs matter
Neural Processing Units are specialized accelerators built to run AI workloads more efficiently than general‑purpose CPUs or GPUs for certain tasks. The 40+ TOPS (trillion operations per second) threshold used by Microsoft is a performance floor intended to ensure a consistent end‑user experience for on‑device models — enabling real‑time or near‑real‑time generation without unacceptable lag or thermal throttling.NPUs are optimized for low-precision matrix math and transformer inference patterns, which lets manufacturers deliver respectable local model performance in mobile laptops and fan-optimized designs. This hardware investment is what makes local models viable in consumer notebooks, moving these features from data center‑bound experiments into everyday apps.
What to watch next
- General availability: The preview is in Canary and Dev channels. Watch for the public release window and the exact OS build requirement.
- Model transparency: Expect pressure for clearer documentation on which local models are used, their size, update cadence, and privacy/telemetry behavior.
- OEM coverage: More vendors will announce Copilot+ SKUs if demand for local AI features grows. Look for detailed hardware compatibility lists.
- Enterprise controls: Microsoft and third‑party MDM vendors will expand controls for model provisioning and telemetry in enterprise editions.
- Developer access: Microsoft’s writing assistance APIs and Edge/Windows model runtimes may be exposed to developers who want to add local generative features to their apps.
Final assessment
Notepad’s new on‑device generative AI is a pragmatic, incremental step that converts a low‑barrier tool into a privacy‑aware, hybrid assistant for supported Windows 11 hardware. The move is strategically sound: it shows how Microsoft intends to balance local capability with cloud scale, reduce subscription friction for basic tasks, and make privacy‑minded AI accessible.There are real limitations — hardware exclusivity, model capability, and update/telemetry transparency are the most important. For users with Copilot+ hardware, the result is likely to be a fast, private, and useful set of writing helpers built into a familiar app. For everyone else, the update is a reminder that not all AI features are created equal: access will increasingly hinge on device class and vendor partnerships.
This is a welcome evolution for Notepad and for Windows as a platform: a move toward local intelligence that respects both utility and privacy while keeping the door open for cloud power when needed.
Source: Windows Central Windows 11 Notepad will soon let you generate text using on-device AI models — no subscription required
