Notepad Gets On-Device AI with Copilot+ — Summarize, Write, Rewrite

Microsoft is rolling AI into the smallest, most ubiquitous text editor on Windows — Notepad — by adding on‑device generative features for Copilot+ PCs, and the changes signal a meaningful shift in how Microsoft blends local and cloud AI across Windows inbox apps.

Overview​

Notepad’s September Insider update (Notepad version 11.2508.28.0) introduces three built‑in generative actions — Summarize, Write, and Rewrite — and for compatible machines these can run completely locally on the device’s Neural Processing Unit (NPU) without requiring a Microsoft account or Microsoft 365 subscription. The update is rolling to Windows Insiders in the Canary and Dev channels first, with broader availability to follow for Windows 11 users on qualifying hardware.
This feature set reframes Notepad from a barebones scratchpad into a lightweight, situational writing assistant intended for quick drafts, tone adjustments, and compressed summaries — all surfaced directly in the editor via the Copilot menu, right‑click context menu, and in some Insider flights, keyboard shortcuts.

---

Background: why this matters​

Notepad has long been the single‑purpose, instant‑open plain‑text tool in Windows. Over the past year Microsoft has steadily modernized inbox apps — adding tabs, spell check, autocorrect, and basic AI helpers — but those earlier AI capabilities typically relied on cloud models and, in many cases, Microsoft account sign‑in plus subscription credit systems. The new Notepad update formalizes a hybrid approach: local inference on Copilot+ hardware for common tasks, with cloud fallback and higher‑capacity models available to subscribers.
There are three strategic implications here:

• A privacy‑centric option: local inference reduces the need to send user text off‑device, which is meaningful for sensitive notes and offline work.
• A performance/economics trade: running models on an NPU aims to lower latency and remove subscription friction for baseline features, while cloud models remain the premium option for larger, more up‑to‑date reasoning.
• A hardware stratification: these benefits are gated to Copilot+ PCs — a device class defined by on‑device NPUs and hardware/firmware certification — which creates a split in user experience across the Windows ecosystem.

---

What’s new in Notepad: the three AI modes​

Summarize​

Summarize compresses selected text into a short, medium, or long summary and can be used to extract action items from meeting notes, logs, or long drafts. The action is designed for speed and to keep users working inside Notepad rather than switching contexts.

Write​

Write can generate new text from a short prompt or expand a selected fragment into paragraphs, lists, or templates — useful for quick email drafts, outlines, or fleshing out terse notes. This is the generative composition mode for when users want to originate content.

Rewrite​

Rewrite rephrases existing selections to change tone, formality, length, or clarity. It returns alternative variants so users can choose the best fit and is aimed at fast polishing rather than deep editorial work.
All three actions are surfaced via Notepad’s Copilot menu and the right‑click context menu; in Insider builds Microsoft is experimenting with keyboard shortcuts and more compact UI flows for these operations.

---

How the on‑device AI works (high level)​

Notepad’s local AI path requires two core elements:

• A Copilot+ PC with an NPU and firmware/drivers that present the hardware capabilities to Windows.
• A local model runtime and model weights that Windows or the OEM provisions to the device (either preinstalled or downloaded by the OS as needed). When Notepad detects a qualifying Copilot+ PC, the app can perform inference locally on the NPU (and, if necessary, the CPU/GPU) so text never needs to leave the device for these specific operations.

Microsoft’s public guidance and accompanying reporting describe 40+ TOPS (tera‑operations per second) as a commonly referenced performance threshold for Copilot+ NPUs — a practical marker for interactive local inference on battery‑powered laptops and tablets. That 40+ TOPS figure appears repeatedly in vendor and Microsoft communications as a baseline for Copilot+ class hardware.

---

Hardware and certification: what “Copilot+ PC” means in practice​

Copilot+ is a device‑level certification tied to silicon, firmware, and driver support. Typical examples of platforms that meet Copilot+ criteria include modern Snapdragon X‑series SoCs and recent AI‑accelerated variants from Intel (Core Ultra) and AMD (Ryzen AI) that expose an NPU to the OS. The Copilot+ designation also implies OEM‑level support — drivers, firmware, and Windows components must be present to enable on‑device model runtimes.
Important practical notes:

• Copilot+ is not a toggle you can flip in Settings; it’s a certified hardware/firmware profile shipped by OEMs.
• Minimum RAM and storage expectations typically place Copilot+ devices in the premium segment (many OEMs target 16 GB RAM and NVMe SSD baselines), though Microsoft’s published consumer guidance focuses on NPU capability rather than rigid memory thresholds.
• If your device lacks Copilot+ certification, Notepad’s AI actions will route to cloud models where applicable — and that path may require sign‑in or a Microsoft 365 subscription for higher‑capacity or specialized cloud models.

---

Availability and rollout​

The Notepad build with AI features is being flighted to Windows Insiders (Canary and Dev channels) under Notepad 11.2508.28.0 and is expected to reach wider Windows 11 users with compatible hardware in the weeks after the Insider preview. Microsoft’s rollout strategy uses staged feature flights and channel gating, so not every Insider will see the capability immediately even on qualifying hardware.
The local model support is initially limited to English only; cloud models remain the path for multilingual support and more up‑to‑date knowledge.

---

Privacy, enterprise governance, and risk profile​

Adding local inference to Notepad reduces cloud exposure for specific operations, but it does not eliminate governance responsibilities. Administrators must consider:

• Data routing: while local inference keeps text on device for supported actions, other Copilot and Visual Search integrations across Windows apps can still send data to Microsoft services depending on which action users select. Audit trails and data flow maps are required to understand where content leaves the device.
• DLP and auditing: enterprises should test how Quick Markup, Visual Search, and Ask Copilot interactions behave with existing DLP/proxy configurations; some actions may change clipboard/autosave semantics.
• Model governance: local models remain capable of hallucinations and factual errors. For regulated or safety‑critical contexts, treat generative outputs as assistive drafts that require human verification. Implement label‑and‑review workflows where AI‑assisted content is used externally.

From a compliance perspective, the local model option is a useful privacy tool but not a silver bullet. Decision makers should pilot on representative Copilot+ devices, map how AI features interact with networked services, and update acceptable‑use policies accordingly.

---

Performance, model details, and transparency — what Microsoft has and hasn’t disclosed​

Microsoft’s Insider posts and supporting documentation confirm the functional behavior — that Notepad’s Summarize/Write/Rewrite features exist, and that on Copilot+ hardware they can run locally without a subscription. However, Microsoft has not publicly published granular model architecture, parameter counts, or complete on‑device performance benchmarks for the local Notepad model. Likewise, precise NPU thresholds beyond the common 40+ TOPS guideline, expected disk footprint for model weights, or RAM usage profiles are not enumerated in the public announcement.
Because those details materially affect provisioning and enterprise risk assessment, treat any claims about parity between local and cloud models, or about specific latency/accuracy tradeoffs, as unverified until Microsoft releases technical documentation or third‑party benchmarks. Expect variability in output quality and speed across different Copilot+ hardware implementations.

---

Practical user guidance: how to try, tune, and (if desired) opt out​

If you want to test Notepad’s AI features as an Insider:

• Join Windows Insider on the Canary or Dev channel and update Notepad via the Microsoft Store or through Inbox app flighting.
• Confirm your device is Copilot+ certified (check OEM documentation or device specifications that reference Copilot+ hardware).
• Open Notepad, select text and use the Copilot menu or right‑click to choose Summarize, Write, or Rewrite. Compare local vs. cloud outputs if you have a Microsoft 365 subscription and the cloud option is available.

For users who prefer the classic Notepad experience, Microsoft provides a setting to disable the AI features inside the app so Notepad remains a no‑frills plain‑text editor. That choice respects users who want the historical simplicity.

---

Strengths and immediate benefits​

• Low‑friction productivity: embedding quick drafting and rewriting inside Notepad reduces context switching for common tasks like drafts, notes, and paraphrasing.
• Privacy‑oriented option: local inference on Copilot+ PCs offers a privacy win for sensitive text and offline workflows.
• Democratizes access on qualifying hardware: baseline generative features are available without subscription on Copilot+ hardware, lowering friction for experimentation.
• Rapid iteration surface: Notepad is a low‑risk surface for Microsoft to test incremental generative features before expanding to heavier workloads.

---

Risks, tradeoffs, and important caveats​

• Hardware segmentation: the Copilot+ gating creates divergent experiences across Windows devices; many users and organizations will not see local AI benefits until hardware is refreshed or upgraded.
• Opaque model specifics: the lack of published model specs and performance benchmarks makes capacity planning and risk assessment difficult for IT teams. Treat claims of parity with cloud models cautiously.
• Hallucination risk: generative outputs can be plausible but incorrect — particularly risky when used for factual summaries or business communications without verification.
• Governance surface area: additional integrations (Visual Search, Ask Copilot, sharing flows) broaden the places where data can leave the device — enterprises must re‑test DLP, clipboard, and autosave workflows.
• Device provisioning and updates: on‑device models require local storage and update/patch processes; organizations should account for update windows, bandwidth, and model lifecycle management. This remains an operational detail Microsoft has not fully disclosed at the time of the Insider flight.

---

Recommendations for IT admins and power users​

• Pilot on a representative set of Copilot+ hardware before broad deployment to measure latency, output quality, and resource impact.
• Update acceptable‑use and DLP policies to map which Notepad actions run locally versus which actions invoke cloud services; log and audit user interactions where compliance requires traceability.
• Treat AI outputs as drafts and set workflows that require human review for external or regulated communications.
• If sensitivity is paramount, validate that the device is operating in local mode before using Notepad’s AI features for confidential content.
• Communicate with device OEMs and Microsoft channels for more detailed technical documentation as it becomes available — model specs and performance guidance will be critical for enterprise adoption.

---

What remains unverified and what to watch next​

The most important unknowns to monitor are:

• Exact on‑device model architecture, parameter count, and runtime resource usage (disk, RAM, NPU cycles).
• Precise NPU performance minima and how OEMs will report Copilot+ certification details per SKU.
• Real‑world performance and quality comparisons between the local Notepad model and Microsoft’s cloud models across varied input types.
• How Microsoft will manage local model updates, telemetry, and opt‑in/opt‑out behavior for organizational policies.

Until Microsoft or independent benchmarkers publish those specifics, treat model parity, latency guarantees, and precise hardware thresholds as unverified. Enterprises and power users should wait for explicit technical documentation or produce empirical results from controlled pilots.

---

Final analysis​

Turning Notepad into a situational AI assistant is an illustrative step in Microsoft’s hybrid AI strategy: ship baseline, privacy‑preserving capabilities on capable hardware while preserving cloud paths for higher capacity and multi‑language needs. This approach delivers tangible benefits — lower latency, offline operation, subscription‑free baseline features for Copilot+ owners — but it also creates a more fragmented experience across Windows devices and raises governance, provisioning, and transparency questions that organizations must address.
For everyday users, the change is mostly positive: Notepad becomes more useful for quick drafting and editing tasks without replacing heavier authoring tools. For enterprises, the update is an operational signal — on‑device AI is real, but responsible adoption requires careful piloting, policy updates, and attention to the gaps that Microsoft has not yet fully documented. Treat the Notepad AI arrival as an invitation to experiment, not a turnkey solution for production‑grade generative workflows.
The update is live for Insiders now; expect broader availability to follow as Microsoft firms up model documentation, OEM certification details, and enterprise controls.

---

Source: Red Hot Cyber Ready for AI-powered Notepad? Coming soon to Windows 11 with PC Copilot+!