Personal Agent Ecosystems 2026: Muse vs Cue vs Dots vs Grok Bot vs Today AI
Late 2026 is when personal AI agents went mainstream: Meta Muse (Sep 8), Manus Cue (Sep 28), OpenAI Dots (Sep 29) and Qi Junyuan's Today AI launched within weeks of each other, with xAI's Grok Bot alongside. This hub compares their architecture, credential handling and autonomy model โ not marketing claims. Only publicly verified facts are stated; unknowns are labeled.
The five agents at a glance
| Agent | Launched | Execution model | Credentials | Deep-dive |
|---|---|---|---|---|
| Meta Muse Meta | Sep 8, 2026 (US/CA) | Isolated Linux "Muse Secure VM" with systemd-nspawn containers | "Sentinel" host process monitors egress via eBPF; "authd" daemon injects surrogate credential tokens | Read โ |
| Cue Manus (independent again) | Sep 28, 2026 (invite) | Client-side daily OS / planner focused on calendar & task state | OAuth integration to Gmail / Outlook / Calendar endpoints | coming soon |
| OpenAI Dots OpenAI | Sep 29, 2026 | Dedicated persistent cloud computer + browser per instance (GPT-6 "Astra") | Read-only background research restrictions until approved; memory persists across runs (no fine-grained delete UI yet) | coming soon |
| Grok Bot xAI / SpaceXAI | Late Sep 2026 | Distributed cloud nodes โ keeps working when your device is off | Cross-device state sync: Desktop, Mobile and xAI backend | coming soon |
| Today AI Cijian Wuqi (้ฝไฟๅ
, ex-Teambition / ex-่ฑๅ
PC) | Jul 2026 beta ยท Sep 2026 overseas | Proactive personal agent with autonomous Memory management โ gathers task-relevant personal info and connects your apps | App connections ("connect apps") for memory sourcing; called the most proactive agent in current reviews | coming soon |
The three architecture patterns
Cue currently sits outside these three: it is client-side (calendar/task planner over OAuth) โ the lightest-weight pattern, and the reason it ships to consumers fastest.
FAQ
What is a personal AI agent (and how is it different from a chatbot)?
A personal agent is an AI that works autonomously toward a goal: it has its own runtime (a virtual computer or sandbox), can hold credentials to your accounts (email, calendar, payments) and executes multi-step tasks in the background โ no prompting required. A chatbot answers; an agent acts. The 2026 wave (Muse, Cue, Dots, Grok Bot, Today AI) made this category mainstream.
Muse vs Cue vs Dots vs Grok Bot โ which should I pick?
They are barely comparable today, because each exposes a different architecture: Muse runs your tasks inside isolated Linux VMs with eBPF-monitored egress and virtual payment cards; Dots gives each instance a persistent cloud computer on GPT-6 Astra; Grok Bot runs on distributed cloud nodes that keep working when your device is off; Cue (invite-only) is a client-side planner over your calendar and email via OAuth. Pick by the integration surface you need, not by marketing.
How do these agents hold my credentials safely?
Muse is the most explicit: an authd daemon injects surrogate tokens instead of raw passwords, and an eBPF program on the host watches network egress; payments go through single-use virtual cards (Stripe Link). OpenAI Dots restricts unapproved instances to read-only research and keeps memory across runs โ note there is no fine-grained memory-deletion UI yet. Cue uses standard OAuth to Gmail/Outlook/Calendar. Grok Bot syncs state across devices via the xAI backend; its proxy encryption specs are not public.
Do any of these agents work while my computer is off?
Grok Bot explicitly markets always-on distributed cloud nodes that continue working when your device is offline. OpenAI Dots gives each instance a persistent cloud computer, so tasks survive locally too. Muse runs inside its Secure VM in Meta's cloud. Cue is currently the most client-side: its daily planner is oriented around your calendar/task state on device.
Building agents yourself? See our JEV decision-model directory, DeepSeek Harness plugins and Codex plugins.