Free, open-source desktop assistant that runs open-weight models — Gemma, gpt-oss — entirely on your own computer, picks the best model for your hardware, and can run fully offline with no account or subscription.
Hob runs open-weight models such as Gemma and gpt-oss entirely on your own computer, picking the best model for your hardware — so your files, PDFs and spreadsheets never leave the machine.
It works on your files, PDFs and spreadsheets, can run fully offline, and needs no account or subscription, with downloads for Mac, Windows and Linux.
| Pricing | Open Source |
|---|---|
| Category | AI Personal Agents |
| Platforms | mac, windows, linux |
| Company | Frigade |
| Best for | Anyone who wants a private AI assistant that never sends their files to a server |
| Official website | hob.chat |
Personal agents go further than chatbots: they keep working in the background, use tools, and complete multi-step tasks on your behalf. Choosing one starts with trust and control. Look for clear permission systems — what the agent may do on its own, what needs your approval, and how you revoke access — because an agent connected to your email, calendar and files is powerful in both directions. Channels matter: an agent you can reach by text, call or inside your chat apps fits real life better than another dashboard. Check memory and continuity (does it remember your preferences and past tasks?), the models it can use, and whether it is hosted for you or self-hosted on your own infrastructure. Start any agent on low-stakes tasks, review its work, and widen its permissions only as it earns trust.
Hob runs open-weight models such as Gemma and gpt-oss entirely on your own computer, picking the best model for your hardware — so your files, PDFs and spreadsheets never leave the machine.
Hob offers On-device open-weight models — Gemma, gpt-oss and more, Automatic model pick matched to your hardware, Works on your files, PDFs and spreadsheets, Fully offline mode with no account required and Mac, Windows and Linux downloads, free with no subscription.
Hob is open source, so the software itself is free to download, use and self-host. Your only costs are the infrastructure you run it on and any paid services you connect to it.
Strengths of Hob include Private by design — AI runs on your computer, not a server; Free with no subscription or account; Works fully offline. On the downside, Performance depends on your own hardware; Open-weight models only — no frontier hosted models.
by JIMMY
AI agent that lives on your Windows PC and takes orders from your own WhatsApp chat: find and send files, open apps, run tasks — with approval before every send.
Best for: Windows users who want to run small PC jobs from a WhatsApp chat
Pricing: Paid
Limitation: Not an official WhatsApp client — WhatsApp's terms do not allow unofficial clients, so the linked number could be restricted or banned (maker's own FAQ warning)
by Smart Jelly
Private personal AI for Mac: chat, voice, dictation, meeting notes, documents and agents joined by one on-device memory of your work.
Best for: Mac users who want one private AI that remembers their work across chat, voice and meetings
Pricing: Freemium
Limitation: Mac only
by Empacer
Voice-first AI assistant and workspace for small business owners: speak instructions out loud and it drafts marketing, inbox replies and documents from its memory of your business.
Best for: Small business owners who know what needs doing but never get around to writing it
Limitation: No pricing shown in the official site text fetched
by Dotverse
Mac app that brings your existing ChatGPT "Dot" to your Mac as a crew of role and personality Dots sharing Dot's memory.
Best for: ChatGPT Dot users on Mac who want one Dot to act as a crew of specialised roles
Limitation: Requires an existing Dot in ChatGPT — it provides no AI service of its own
by Kiwi
Proactive AI assistant that texts you first — it reads calendar and email, texts once at the right moment, and acts on any app or website on your yes, in iMessage or WhatsApp.
Best for: People who want an assistant that pings them at the right moment instead of waiting to be asked
Pricing: Free
Limitation: Free during the pilot — pricing after the pilot is unannounced
by llama.cpp
Tiny menu-bar and system-tray app that runs open models locally via llama.cpp — chat with them, or point any OpenAI-compatible app at its local API, with one copy of each model shared across apps
Best for: People who want to run open models locally and share them across chat, coding agents and their own apps
Pricing: Open Source
