Rabbit is launching OS3, an Artificial Intelligence agent (AI agent) for Windows, Mac, and Linux computers that does not require the company’s poorly received R1 device. At the same time, narrower assistants are emerging for voice input, website creation, home searches, and business processes. For you, the useful distinction is becoming less about which chatbot writes best and more about which tasks you can safely hand over.
Why AI agents are moving closer to everyday use
An AI agent is software that does more than discuss a goal: it selects and carries out several steps toward completing it. Rabbit’s new OS3 system is supposed to determine which devices, files, applications, and AI models a task requires, according to the provider. One account can connect up to five devices, with access available through a dedicated desktop site, paired messaging services such as Telegram or iMessage, and the R1 device.
Rabbit calls OS3 an “agentic operating system,” although the service runs in the cloud while operating local computers. That distinction matters: performing actions on your computer does not mean all associated data stays there. Rabbit says it will not store, copy, use, or sell your data, yet it will retain chats and memories derived from them on its servers; connected AI providers will also process information under their own privacy rules.
The company’s hardware strategy is changing as well. Rabbit founder Jesse Lyu reportedly said that production of the R1 has stopped. Rabbit is instead working on a “cyberdeck” for vibe coding, meaning software creation through natural-language instructions, that is expected to run OS3 when it arrives within months. The report gives no price, firm release date, or independent assessment of the promised capabilities.
What specialized everyday tools can do
Not every assistant is trying to control your entire computer. According to a summary of the SpeakON voice button, the device tackles a narrower problem: conventional dictation often captures filler words and false starts, leaving you to clean up the text and move it into another application. The MagSafe button, which has its own microphone, is intended to turn speech into polished communication and actions across apps.
One practical use would be dictating a message while away from your desk and receiving something more usable than a raw transcript full of “ums” and abandoned sentences. However, the source provides neither a price nor details about Swiss availability, and only a short summary is available. Its performance with Swiss German, accents, or confidential conversations therefore remains unclear and has not been independently verified.
Swiss hosting provider Hostpoint now offers an AI assistant that creates an initial website structure from information about a company, its industry, and the site’s purpose. It proposes a page structure, design, text, and images. The result uses WordPress, a content management system (CMS) that lets you manage content without building every component from scratch, and can later be edited with themes, plug-ins, and standard WordPress functions.
According to Hostpoint, the assistant is available at no extra charge to customers with Smart or Business web hosting or a Managed Flex Server. It can also help with later revisions. For a small Swiss business, a realistic example would be generating an initial company site with a service overview, contact page, and proposed visual design; checking factual claims, wording, and image rights remains a human responsibility.
Comparis is bringing the home search directly into ChatGPT. Its plug-in, an installable extension, lets you request something as specific as a three-room apartment in Zurich with a kitchen island and wooden floors. Comparis says the tool can analyze listing photos to identify features that are unavailable as standard text filters, and a listing discovered through ChatGPT can be opened in the Comparis real estate app.
What agents change at work
Workplace uses range from scientific papers to warehouse operations. Stanford’s Paper2Agent makes individual research papers interactive: the resulting agents are designed to answer questions about a paper, validate results, and communicate with other agents. The framework, a technical toolkit, is available at no charge, but it requires code and a coding assistant such as Claude Code, so it is not a typical beginner product despite being freely available.
For a student or professional, such an agent could make it easier to ask targeted questions about a study’s method and findings instead of reading only from beginning to end. It does not replace reviewing the original paper or applying subject-matter judgment. When a tool claims to “validate” results, you also need to know whether it is reproducing calculations or merely producing plausible language; the source supplies no independent quality measurement on that point.
AutoScheduler takes an operational approach with its warehouse application builder. Logistics teams are meant to create custom tools directly from current facility data. The module is part of a broader Warehouse AI Platform for distribution centers that must coordinate inventory, machinery, and labor, and is positioned as a more flexible complement to rigid enterprise planning and warehouse management systems.
An internal chatbot at the Rewe Group illustrates why technical performance alone is insufficient. It answers questions using existing internal knowledge, accesses material in Confluence and SharePoint, and creates a support ticket when necessary; for privacy and compliance, meaning adherence to organizational requirements, it had to operate permanently in Europe. Even so, employees initially made little use of the technically functional “Hotline Agent.”
Renaming it “Lumi,” adding an avatar, and running an internal brand campaign increased usage, according to Rewe’s account of AI adoption. The system now reportedly resolves 5,000 requests automatically each month and serves roughly 20,000 administrative employees. A Bitkom study cited in the report fits that experience: 31 percent of companies identify weak employee acceptance as one of their largest obstacles to adopting AI.
Pros and Cons of AI agents in everyday life
Pros:
- Fewer handoffs – Agents can find information, prepare content, and initiate actions in connected applications.
- More natural input – You can describe apartment criteria, website requirements, or support problems in ordinary sentences instead of rigid menus.
- Focused utility – Narrow tools such as listing-image searches and warehouse app builders address identifiable tasks.
- Lower entry barrier – Existing Hostpoint customers receive the website assistant at no extra charge, according to the provider, while Paper2Agent is available for free.
Cons:
- Data spread across services – With OS3, Rabbit and connected AI providers may each apply separate rules to chats, memories, and other information.
- Unverified reliability – Several performance claims come directly from providers and were not independently tested in the supplied reports.
- Unclear total cost – The available information gives no complete pricing for OS3, SpeakON, or various supporting services that may be required.
- Adoption requires work – Rewe’s experience shows that even a functional agent may go unused without a clear introduction and employee trust.
What this means for you in Switzerland
If you are a beginner, start with a narrow task whose output is easy to check. If you already have an eligible Hostpoint plan, for example, you could generate a basic website and review every page before publication; for a home search, you could install the Comparis plug-in in the ChatGPT app or browser and test a specific combination such as room count, location, kitchen island, and wooden floors. Both services have a direct Swiss connection, but generated content and retrieved listings are starting points rather than verified results.
If you are an advanced user, you can get more value by separating tasks, data access, and review points. With OS3, that could mean connecting only the devices and AI models required for a job and checking in advance which chats or memories remain on which servers. In an organization, a restricted use case based on existing knowledge, a clear handoff to human support, and a comprehensible internal rollout follow the lesson of Rewe’s experience.
For Swiss users, Hostpoint and Comparis are the most immediately relevant services in this group. The sources do not provide complete information about Swiss German performance, every service’s data location, or pricing for the international products. Voice input, internal documents, and personal search profiles are especially good reasons not to confuse “local operation” with completely local data processing.
These products show a credible shift from chatbots that answer toward assistants that act, but the case is stronger for tightly bounded tasks than for universal agents. Website drafts, detailed home searches, and internal support requests are easier to review than a system with access to multiple devices, files, and models. The unresolved risk is that convenient automation may demand more data access and trust than the available evidence currently justifies.
Sources
- Rabbit’s new AI agent doesn’t need an R1 to run – Unknown publication, 2026-09-22
- SpeakON Ships a MagSafe AI Voice Button With Its Own Microphone – Unknown publication, 2026-09-23
- Hostpoint lanciert KI-Website-Assistenten – Unknown publication, 2026-09-22
- Comparis lanciert ChatGPT-Plug-in für die Wohnungssuche – Unknown publication, 2026-09-22
- Studie als KI-Agent: Stanford-Tool macht Forschungsartikel interaktiv – Unknown publication, 2026-09-22
- AutoScheduler launches warehouse app builder for logistics teams – Unknown publication, 2026-09-22
- KI-Akzeptanz: Wie Rewe digital einfach nur den Chatbot umbenannte – Unknown publication, 2026-09-22
Image: Jeremy Enns via Pexels