Appendix: Personal Agent Software¶

Research notes — not yet a finished chapter

This appendix holds raw research gathered on October 15, 2026. Nearly every claim is vendor-stated or comes from press coverage; we found no independent benchmarks or controlled tests of any product listed here. Prices, tiers, and features changed several times within weeks of launch, so verify against each vendor's own pages before relying on a number.

For a one-page picture of the four commercial products, see the interactive poster Four Personal Agents, Side by Side.

Why this matters for EE research¶

A personal agent is an AI system that does not just answer a question but carries out multi-step work on your behalf — signing in to tools, running routines, and coming back when it is done or needs a decision. For a researcher, that could mean overnight literature sweeps, rerunning an analysis when new lab data lands, or filing bug tickets against a simulation toolchain. The products below all arrived within roughly two months of each other in mid-2026, so the category is moving fast.

Grok Bot (SpaceXAI)¶

Grok Bot app icon

SpaceXAI announced Grok Bot on August 11, 2026, as an early-beta, cloud-based agent service. "Bot" is the product name; it is not the ordinary Grok chatbot. It appears to be the first joint product of xAI and Cursor. Sources disagree on whether SpaceX is acquiring Cursor or merging with it (MacRumors cites a $60 billion acquisition), so treat the deal terms as unsettled.

Capabilities¶

Persistent cloud computer. Each user gets a cloud machine with a browser, files, and a terminal. Work continues when the user's laptop is off.
Signs in like a human. Bots log in to websites and apps, including services that offer no API or MCP integration.
Teachable routines. A user shows a bot a task once; it saves the routine and later runs it on its own.
Multiple bots. Several bots can share one group chat, share context, and hand work to each other. A manager bot can delegate to specialist bots. AIMultiple reports up to six bots per group chat and "Team Bots" on Teams/Enterprise plans (public beta since September 28).
Memory. Bots keep memory across conversations and may start work "before you need to ask," per SpaceXAI.
Interface. Work is assigned in a chat window on desktop or iOS. The bot returns a result or asks for approval.

SpaceXAI's own example uses include updating a CRM from call transcripts, onboarding new hires and processing invoices in Gmail, and reproducing a bug, filing a ticket, and handing the fix to a debugging bot.

Access, pricing, and platforms¶

Item	Detail
Launch access (Aug 11)	SuperGrok Heavy, Cursor Ultra, Cursor Teams Premium
Expanded access (Aug 26)	SuperGrok, SuperGrok Plus, SuperGrok Heavy, Cursor Pro, Pro+, Ultra, Teams Standard and Premium
Launch prices	About $300/mo (SuperGrok Heavy), $200/mo (Cursor Ultra), $120/seat/mo (Teams Premium)
Entry prices (Oct 12)	About $30/mo (SuperGrok), $20/mo (Cursor), $40/seat/mo (Cursor Teams)
Usage	Weekly allowance included; overage billed at token cost; allowance and rate card unpublished
Platforms	macOS, Windows, Linux (.deb), iOS; Android "coming soon"
Models	Not disclosed; AIMultiple reports Cursor picks a model per task
Enterprise	Waitlist at launch; "through sales" by October 12

Security claims and open questions¶

SpaceXAI states that Grok Bot uses Cursor SSO and authentication, privacy mode, encryption in transit and at rest, a training opt-out, and an "Auto Review" step for sensitive actions. Enterprise admins can configure data-loss prevention, certificates, proxies, and network controls.

Open questions raised by analysts:

Shared logins. Isolation is per user, not per bot. All of a user's bots share one computer, including its files, browser, and logins, so one compromised bot exposes the rest.
Undocumented details. Retention defaults, credential handling, permission granularity, audit records, which actions need approval by default, and whether memory can be viewed or edited.
Reliability. Nothing documents failure handling or what happens to a saved routine when a website's interface changes.
Cost surprises. Cursor itself warns that an hourly routine or a busy Slack trigger "can use a week of usage in a day."

Dots (OpenAI)¶

ChatGPT app icon, the app that hosts Dots

OpenAI introduced Dots at DevDay on September 29, 2026, as a personal agent inside ChatGPT, powered by GPT-6 Astra.

Always-on. A dot runs in the background with its own cloud computer and browser, pursuing goals with minimal oversight.
App ecosystem. It can use the 4,000+ apps already connected to ChatGPT and can delegate coding to Codex.
Visibility. The user can open a dot's "computer" at any time to see what it is doing, and can connect a dot to their own PC.
Communication. A dot can message or call the user with progress, questions, or decisions. Slack and Microsoft Teams integrations were announced; text-message support is described as coming later.
Safety model. Passwords used for sign-in are not exposed to the underlying model. Proactive background research uses read-only app access. Dots have built-in rules for when to act versus ask, and users can write custom approval rules. OpenAI says it is working with Microsoft to bring Dots under Agent 365 security controls.
One dot for now. Each eligible user gets one primary dot they can name and customize. OpenAI says it expects "teams of Dots working together," but multi-dot control is not available at launch.
Availability. Launched for Pro and Business Premium users in "eligible markets"; reports differ on whether Business and Enterprise are included, and some exclude the EEA, Switzerland, and the UK for personal accounts.

Conflicting price reports

Engadget reports the first dot is free for Pro and Business Premium subscribers. AIMultiple lists a $100/mo starting price. We could not reconcile the two; check OpenAI's own announcement.

TechCrunch notes that much of what Dots do was already possible with Codex and similar agent harnesses — Dots package those capabilities around independent action.

Muse (Meta)¶

Muse from Meta app icon

Meta launched Muse on September 8, 2026, in the U.S. for users 18 and older, as a separate app that also runs inside WhatsApp. It is aimed at everyday tasks: schedules, shopping, travel, and turning long-term goals into plans.

Architecture. Each user's agent runs in its own cloud virtual machine.
Permissions. A layer called Sentinel decides whether each connected-app action and outbound request is allowed, blocked, or sent for approval. Meta says the agent never sees your actual passwords or payment details, and each purchase needs approval showing exact details.
Visibility. Users can watch the agent's browser as it works and review an activity log.
Memory. Stored as editable text files.
Pricing. Free tier, plus $20/mo (Power) and $100/mo (Maximum) tiers with weekly token allowances.
Privacy. Training on user data is on by default, with an opt-out.
Integrations. Centered on Meta's platforms plus email, calendar, payments, and a handful of consumer categories — narrower than the developer-oriented tools.

Claude Cowork (Anthropic)¶

Claude app icon

Cowork launched in January 2026 and became generally available on April 9, 2026. It lives in a tab inside Claude Desktop, where Claude reads your files, runs multi-step tasks, and uses plugins and MCP connectors.

Platforms. macOS first, with Windows following in February 2026. Anthropic's download page also lists Linux and ChromeOS builds.
Scheduled tasks. Added February 25, 2026. A March 2026 source says they require the computer to be awake and Claude Desktop open, which makes Cowork the least "always-on" of the commercial products. That may have changed since.
Dispatch. Lets you assign a task from your phone.
Pricing. No separate SKU: it is included in Claude Pro ($20/mo), Max ($100 or $200/mo), Team, and Enterprise plans. Agentic tasks reportedly burn usage allowances five to twenty times faster than ordinary chat, and the usage pool is shared with Claude chat and Claude Code.
Enterprise. Role-based access control, per-group spend limits, and usage analytics. One source notes Cowork activity is not yet captured in audit logs or the Compliance API.

Open-source alternatives: OpenClaw and Hermes Agent¶

OpenClaw lobster mascot

OpenClaw is an MIT-licensed, self-hosted agent that lives in chat apps such as WhatsApp, Telegram, and Discord. You choose the model (Claude, GPT, Gemini, or a local model through Ollama). Model usage is the main cost — one estimate is $5–50/month. Reviews report reliability problems: runaway loops, malicious skills, and unattended deletions.
Hermes Agent (Nous Research, public since February 25, 2026) is also MIT-licensed. It runs on a laptop, server, or cloud sandbox, supports many model providers, and has a learning loop that writes its own skills. The trade-off is that you manage your own servers, secrets, and security.

Press describes Grok Bot as "a user friendly OpenClaw": the same idea, hosted and managed, in exchange for openness and control.

Side-by-side comparison¶

	Grok Bot	Dots	Muse	Claude Cowork	OpenClaw / Hermes
Vendor	SpaceXAI / Cursor	OpenAI	Meta	Anthropic	Community / Nous Research
Announced	Aug 11, 2026	Sep 29, 2026	Sep 8, 2026	GA Apr 9, 2026	Hermes public Feb 2026
Runs where	Cloud computer shared by your bots	Own cloud computer and browser	Dedicated cloud VM per user	Mostly local sandbox plus connectors	Your laptop, server, or cloud
Model	Undisclosed	GPT-6 Astra	Muse Spark	Claude	Any
Multi-agent	Yes: many bots, group chats	One dot now; teams planned	One agent	Single session	Build your own
Always-on	Yes	Yes	Yes	Needs computer awake (Mar 2026 source)	Yes, if you host it
Integrations	Browser/computer use, Slack, connected apps	4,000+ ChatGPT apps, Codex, Slack/Teams	Email, calendar, payments, WhatsApp	MCP connectors, plugins	Chat apps, skills, your own tools
Safety model	Auto Review model; shared logins	Read-only background access; approval rules	Sentinel layer; per-purchase approval; activity log	Local sandbox; enterprise RBAC	You manage it
Price	From $20/mo plus usage	Conflicting reports	Free, $20, $100/mo	Included in $20–$200/mo plans	Free software plus API costs
Open source	No	No	No	No	Yes (MIT)

Contrast: what actually differs¶

Grok Bot's differentiator is the manager/specialist bot model. OpenAI says multi-dot control is planned but not shipped; Muse is one agent per user.
Grok Bot is the least transparent. Models are undisclosed, usage allowances are unpublished, and the shared-login design widens the blast radius of any one failure. Dots and Muse document their approval rules more explicitly.
Dots lean on the ChatGPT ecosystem — its apps, Codex, and voice calls.
Muse targets consumers, with the strongest purchase-approval story but the narrowest work-tool integrations.
Cowork is session-oriented and the most mature knowledge-work tool, but the least autonomous.
OpenClaw and Hermes trade convenience for control: you pick the model and own the security.

Caveats and gaps¶

No independent testing. AIMultiple's comparison reports no benchmarks or controlled tests of Dots, Grok Bot, or Muse; the only first-hand evidence is the author's anecdotal use of Grok Bot.
Moving prices. Tiers and prices changed within weeks of each launch.
Beta status. SpaceXAI's October 12 pages checked by Kingy.ai do not say whether Grok Bot is still in beta.
Source bias. Many comparison blogs host or sell one of the products they rank.
Unread source. The Neowin article returned HTTP 403 and was not read directly.
Cowork currency. The newest Cowork source we found is from August 2026.

Questions to resolve before this becomes a chapter¶

Which of these agents suits an EE research workflow (long simulations, licensed EDA tools, lab data) — and which cannot reach those tools at all?
What is each agent's behavior on sensitive or unpublished research data (training defaults, retention, memory)?
How should a research group audit what an always-on agent did overnight?
Which claims can the author verify first-hand with a hands-on trial?

Sources¶

Logos are the vendors' own app icons or project artwork and belong to their owners. The Muse icon is the App Store artwork for Muse from Meta, the OpenClaw icon is from openclaw.ai, and the Hermes Agent wordmark is the banner from the Hermes Agent repository. The Grok Bot, ChatGPT, and Claude icons are cropped from a screenshot of the author's dock. Dots has no icon of its own that we found, so the ChatGPT icon stands in for the app that hosts it.