Notepad adds offline AI on Copilot+ PCs: Summarize, Write, Rewrite

Microsoft has started shipping a Notepad update for Windows 11 that brings true on‑device generative AI — including Summarize, Write, and Rewrite — to Copilot+ PCs, and those capabilities can run entirely offline without a Microsoft account or a paid subscription when executed locally on qualifying hardware. (blogs.windows.com)

Background​

Notepad’s evolution over the last two years has been gradual but decisive: a humble text editor has grown into a lightweight writing assistant layered with Copilot-style tools. Early AI helpers in Notepad required cloud inference, a Microsoft account, and occasionally Microsoft 365 credits. The September Insider update formalizes a hybrid approach: local inference on Copilot+ devices for everyday drafting and editing tasks, with an option to switch to cloud models for higher capacity or multi‑language needs. (blogs.windows.com)
Microsoft’s official Windows Insider post announcing the September rollout lists Notepad version 11.2508.28.0 and explicitly states the new writing tools and the ability to run them locally on Copilot+ PCs without subscription requirements. The post is authored by Dave Grochocki, Principal Group Product Manager for Windows Inbox Apps. (blogs.windows.com)

---

What’s included: Summarize, Write, Rewrite​

Notepad’s new toolkit is deliberately small and task‑focused. Each tool is optimized for quick editing workflows rather than long‑form or production publishing:

• Write — Generate new content from a prompt or expand an existing selection. Useful for drafting emails, filling in notes, or expanding bullet lists.
• Rewrite — Rephrase the selected text, change tone, shorten or lengthen passages, and present alternative variants so users can pick the best fit.
• Summarize — Compress long passages into short, medium, or long summaries or extract key action items from meeting notes and logs.

These actions are surfaced in Notepad via the Copilot menu, a right‑click context menu, and keyboard shortcuts where implemented in Insider builds. The goal is fast, in‑place assistance that keeps the user in the editor rather than forcing app or browser switches. (windowscentral.com)

---

How the offline/local AI works​

The offline capability relies on a two‑part configuration:

• A local language model (model runtime and weights) pre‑provisioned or downloaded to the device by Windows/Microsoft or the OEM.
• A device designated as a Copilot+ PC, which includes a Neural Processing Unit (NPU) able to perform on‑device inference at scale.

When Notepad detects a Copilot+ PC with a suitable NPU, it can run inference locally on the device’s hardware. That means user text does not need to leave the machine for these operations — inference happens on the NPU (and CPU/GPU as required). For users who are not signed in or who do not have a Microsoft 365 subscription, Notepad will fall back to the local model for these features. Subscribed users can choose between the local model and cloud models depending on their needs. (blogs.windows.com)

Copilot+ PC: hardware requirements and what “40+ TOPS” means​

Microsoft’s documentation and Copilot+ PC guidance indicate that many on‑device AI features are gated to devices with NPUs capable of running 40+ TOPS (tera‑operations per second). This is not an arbitrary marketing number — it’s a practical performance threshold for running optimized local models at responsive speed on battery‑powered hardware. Examples of OEM platforms and silicon families that meet Copilot+ criteria include modern Snapdragon X-series, chips branded for AI acceleration from Intel (Core Ultra) and AMD (Ryzen AI) generations, and new purpose‑built NPUs in certain laptops and tablets. (learn.microsoft.com) (prod.support.services.microsoft.com)
Important practical points:

• Copilot+ certification is device‑level — it’s not just a Windows setting. Your laptop/tablet OEM must ship drivers and firmware that present the NPU to Windows as Copilot+ capable.
• Minimum system baselines commonly paired with Copilot+ features in documentation include 256 GB storage and modern CPU/GPU combinations; real‑world performance may favor 16 GB RAM or more depending on model runtime size. (prod.support.services.microsoft.com)

---

Account, subscription, and language coverage​

This update changes a long‑standing assumption: local AI on Copilot+ hardware does not require a Microsoft account or a Microsoft 365 subscription for the local inference path. The Windows Insider announcement explicitly states that users who are not signed in or who lack a subscription can still use the local model for Summarize, Write, and Rewrite; subscribers retain the flexibility to toggle between local and cloud models. At launch, Microsoft notes local Notepad AI supports English only. (blogs.windows.com) (windowscentral.com)
A couple of caveats:

• Cloud models (the higher-capacity or up‑to‑date models that can access web knowledge) remain subject to Microsoft’s cloud account/credit policies and may require sign‑in or a Microsoft 365 plan depending on feature gating and regional policies.
• Local models are intentionally smaller and optimized for offline responsiveness and privacy; they will not match the cloud models in knowledge freshness or necessarily in raw capability for complex tasks.

---

Availability and rollout​

Microsoft is rolling these changes through the Windows Insider Program initially, specifically to users in the Canary and Dev channels. The Insider blog post that announced the update is dated September 17, 2025, and lists the app version numbers (Notepad 11.2508.28.0 among them). Expect a staged rollout: not all Insiders will see the features at once, and general public availability is typically staggered after additional testing. (blogs.windows.com)
Practical testing guidance for Insiders and IT teams:

• Verify the device is listed as Copilot+ certified or meets the NPU/driver requirements.
• Install the updated Notepad app from the Microsoft Store or wait for the Windows Update/Insider flight.
• Test the Summarize/Write/Rewrite flows in local vs. cloud mode and record latency, thermal, and battery impacts.
• Use the Feedback Hub (Apps > Notepad) to report issues and behavior differences across hardware.

---

Privacy, security, and governance implications​

On‑device AI changes the privacy calculus in beneficial ways, but it also introduces new operational considerations.

• Reduced cloud exposure: When inference runs locally, user text does not need to be sent to Microsoft’s cloud for processing, which reduces the attack surface for in‑transit data exposure and may ease compliance in regulated environments.
• Model provisioning and updates: Even when inference is local, the model binaries and runtimes are typically provisioned, updated, and maintained by Microsoft and/or the OEM. Those update channels may involve downloads and telemetry; enterprise administrators should review update, telemetry, and provisioning policies.
• Data egress surface: Notepad’s local mode reduces automatic egress, but Notepad still exposes cloud options for users who switch. Admins must control which mode is allowed using management tooling and policy.
• Auditability and model transparency: Microsoft’s announcement confirms local execution but does not publish granular model specs (model family, parameter counts, or exact resource usage). That opacity makes it harder for governance teams to fully assess hallucination risks, bias characteristics, or forensic auditing of model outputs. Treat claims about parity with cloud models as unverified until Microsoft publishes technical documentation. (windowscentral.com)

Enterprise recommendations:

• Pilot on representative Copilot+ hardware and measure CPU/NPU utilization, latency, thermal behavior, and battery drain during typical Notepad AI tasks.
• Enforce Data Loss Prevention (DLP) and eDiscovery policies that specify allowed cloud/local modes.
• Require human review and labeling of AI‑generated content in regulated outputs; do not rely solely on the model for compliance‑sensitive text.
• Decide whether to enable or block model provisioning via managed update channels depending on policy.

---

Performance, battery life, and practical limits​

Running generative inference on a mobile NPU is a balancing act between responsiveness, model size, and power consumption. Because Microsoft has not published model sizes for the Notepad local models, practical expectations should be measured empirically:

• Expect local inference to be faster (lower latency) than cloud roundtrips for short prompts and editing tasks because it avoids network overhead.
• Expect some battery and thermal impact during sustained use; perform testing under representative workloads before broadly enabling the feature for roaming laptop fleets.
• Expect local models to be tuned for small memory/compute footprints; larger context windows, heavy multi‑document summarization, or complex reasoning tasks may remain more suitable for cloud models. (learn.microsoft.com)

---

How to try it (Insider quick start)​

• Join the Windows Insider Program (Canary or Dev channel) and install the latest app updates.
• Confirm your device is Copilot+ certified (check system/OS build notes or OEM documentation).
• Open Notepad, select text, or position the cursor and open the Copilot menu or right‑click to choose Summarize, Write, or Rewrite.
• If available, toggle between Local model and Cloud model within the Copilot menu to compare output, latency, and behavior.

Note: The exact UI labels and keyboard shortcuts are flight‑dependent and may change between Insider builds and public releases. Microsoft asks Insiders to file feedback in the Feedback Hub (Apps > Notepad) when they encounter problems. (blogs.windows.com)

---

What’s verified and what remains unverified​

Verified (cross‑checked against Microsoft and independent reporting):

• Notepad now includes Summarize, Write, and Rewrite and the update has been shipped to Canary and Dev Insider channels. (blogs.windows.com) (theverge.com)
• Microsoft explicitly supports local model execution on Copilot+ PCs, and local use does not require signing in or a Microsoft 365 subscription. (blogs.windows.com) (windowscentral.com)
• Copilot+ PC hardware guidance requires NPUs capable of 40+ TOPS, and Microsoft’s Copilot+ documentation lists device families and minimum baselines (storage, firmware, etc.) that are commonly associated with these features. (learn.microsoft.com) (prod.support.services.microsoft.com)

Unverified or opaque:

• Microsoft has not published granular details about the local model architecture, parameter size, or exact runtime resource usage for Notepad’s local models. Any claims of exact model parity with cloud models are therefore unverified. Administrators should treat performance and accuracy statements as empirical until Microsoft releases model documentation.

Also note a small but telling discrepancy seen in secondary reporting: one article misspelled the Microsoft product manager’s name as “Dave Grochoki.” Microsoft’s Windows Insider announcement shows the correct name spelling as Dave Grochocki; minor errors like this in aggregation posts underscore why cross‑checking the primary source matters. (blogs.windows.com)

---

Risk analysis and critical considerations​

• Hallucination risk — Even optimized local models can produce plausible but incorrect text. For professional or regulated outputs, always require human verification and introduce labeling for AI‑assisted content.
• Update trust surface — Model binaries and runtimes are delivered through Microsoft/OEM channels; ensure update integrity checks and patch controls are in place to avoid accidental or malicious model swaps.
• Policy fragmentation — Different Windows Insider channels, device certifications, and account states will create a fragmented user experience. Enterprises must standardize which builds and devices are allowed to mitigate surprise behavior.
• DLP and telemetry — While local inference reduces automatic egress, Notepad retains cloud switching and other apps (e.g., Snipping Tool’s Visual Search) that explicitly use cloud services; organizations must update DLP rules accordingly.
• Model auditability — Without published model specs or an audit trail of model versions used for inference, it’s difficult to retroactively investigate why a generated piece of text contained an error. Encourage Microsoft to publish model versioning and change logs for enterprise transparency.

---

How this fits in the broader landscape​

Microsoft’s move follows a broader industry trend toward hybrid models: smaller, on‑device models for routine, privacy‑sensitive tasks, and larger cloud models for heavy lifting. Apple, Google, and open‑source projects have all pursued on‑device AI to varying degrees. By enabling offline Notepad features on Copilot+ hardware, Microsoft is carving a product differentiation path tied to device capability (Copilot+ certification) rather than pure subscription status. That has strategic product implications:

• It rewards OEMs and Microsoft’s device partners who ship capable hardware.
• It reduces friction for basic AI features among users who do not want cloud integration.
• It preserves premium cloud revenue pathways for heavier or cross‑device tasks. (windowscentral.com)

---

Practical recommendations​

For regular users:

• Try Notepad’s Summarize and Write features for note drafting, email stubs, and meeting takeaways — treat outputs as first drafts and edit before sending.
• If privacy is a concern, verify the Copilot mode setting and prefer local inference when available.

For power users and developers:

• Test local vs. cloud outputs systematically and collect examples to judge which mode best fits your workflow.
• Measure performance: latency, NPU/CPU utilization, and thermal impact vary significantly by hardware.

For IT and security teams:

• Inventory devices and identify which are Copilot+ certified.
• Create a pilot group to evaluate DLP impacts and model provisioning behavior.
• Update endpoint policies to control whether local models can be provisioned or whether users are allowed to switch to cloud inference.

---

Conclusion​

Notepad’s new offline AI features represent a meaningful milestone in the shift from cloud‑first to capability‑based AI on Windows. By enabling Summarize, Write, and Rewrite to run locally on Copilot+ PCs without requiring a Microsoft account or subscription, Microsoft has lowered the friction for everyday generative tasks while offering enterprises a more private processing path. That said, the change introduces new governance responsibilities — from provisioning and update controls to auditability and DLP — and administrators should treat the rollout as a testable capability rather than a finished product. The update is rolling to Canary and Dev Insiders now; broader availability will follow as Microsoft iterates on performance, model transparency, and enterprise controls. (blogs.windows.com) (windowscentral.com)

---

Source: Technetbook Windows 11 Notepad to Get Offline AI Writing Features No Internet or Subscription Needed