Newton is a free local AI assistant whose official description names Google's open-weight Gemma model running on the device. It supports questions, writing and brainstorming while keeping conversations in one place. The important boundary is explicit: chat works offline after the app is downloaded, while current-information requests can use DuckDuckGo web search.
That makes local generation and online retrieval different parts of the product, not evidence that every interaction stays offline. This First look is based on the App Store listing. We have not run a prompt, audited network activity or tested performance on a supported phone or tablet.
Based on official store information. We have not tested this app hands-on.
What it does
The basic workflow is opening the app and asking without a separate account or stated usage limit. For a professional, a useful first task is a bounded draft whose requirements can be checked independently, such as reorganizing a non-sensitive outline. The model's response should remain a proposal rather than a verified fact simply because inference occurs locally.
Writing and brainstorming are described at a broad level. The source does not establish document ingestion, access to private mail, calendar actions, voice input or an autonomous tool-execution system. Do not infer those features from the assistant label. Keep the input and intended output narrow enough to see whether Newton actually helps.
Web search through DuckDuckGo supplies a route to more current information when needed. That is a network-dependent operation, and retrieved material still needs evaluation. We have not tested when search is triggered, what query is sent, which sources are selected or whether every answer clearly distinguishes retrieved facts from model-generated wording.
Device support is unusually specific. The description lists selected iPhone generations and variants, with examples including the Pro models in older generations rather than every model carrying a similar number. It also names particular iPads, including A16, A17 Pro and M1-or-later families. Check the exact current compatibility list and system requirements instead of assuming that every device capable of installing the app will offer the intended local-model experience.
Before you install
Local processing changes the data path without making an answer automatically accurate or safe to rely on. A model can produce an unsupported name, calculation or explanation without contacting a server. Verify consequential facts through suitable sources, and do not treat a plausible reply as proof of professional expertise.
The source calls Gemma open-weight, which should not be silently broadened into a claim about every licensing condition, training-data detail or the source code of the whole application. We have not audited the model build or implementation. The named model is part of the developer's description rather than an independent certification.
The offline statement applies after downloading the app. The reviewed text does not explain every installation, update or model-storage condition. Check the actual device setup before depending on a no-signal workflow, and do not invent a separate initial download size or guarantee that every future update works without connectivity.
Online retrieval also deserves its own privacy review. A claim that local chat runs on the device does not establish that a search request is never transmitted. Review the current policy before using confidential client details in a request that may need the web. We have not examined query handling, retention or the completeness of the conversation-recovery process.
Practical usefulness depends on the device and task. We have not benchmarked response speed, battery impact, memory use or long-conversation behavior. A no-usage-limit statement is an access claim, not proof of unlimited hardware resources or a fixed turnaround time. Keep a manageable sample before transferring a larger workflow.
A useful drafting sample can separate the request into facts supplied by the user and wording the model is asked to propose. Check that the response preserves those supplied facts and does not silently add another claim. That makes the result easier to evaluate than an open-ended question whose correctness depends entirely on the assistant's own explanation.
Price and access
The linked US App Store offer is free. The developer describes no account and no usage limits, without naming a paid app tier in the reviewed text. No subscription price, provider allowance or hidden premium boundary is invented here.
Device compatibility, storage and network access for web search remain separate practical requirements. The source does not establish that a free app supplies every external resource without cost or that every answer works offline. We have not tested purchase screens, regional differences or access after an update. Confirm the current offer and the exact supported hardware before relying on it.
Our take
Newton's strongest proposition is a straightforward local drafting and question interface with a separately described path to web search. That may suit a solo professional who wants to inspect a small AI task without creating another service account, provided local processing is not confused with verified reasoning.
Begin with a supported device, a non-sensitive prompt and success criteria that can be checked without the model. Investigate the search boundary separately when current information matters. Our take remains source-based: model quality, privacy implementation and practical performance have not been independently tested, and a local answer still needs an informed human review.
Sources
App icon from the official store listing.
