Claude Duet vs Grok Bot: open source alternative

Claude Duet arrived on October 21, 2026, when Anthropic opened it as a research preview. That puts it alongside Meta’s Muse, OpenAI’s Dots, Hark Pro, and xAI’s Grok Bot in the growing group of hosted personal agents. Each of them keeps working after you close the app. Where they split is the number of agents you get, who operates the computer they use, and whether the model is your call. This Claude Duet vs Grok Bot comparison covers both, then shows where the open source alternative, Krypton, fits.

What is Claude Duet?

Duet is its own app on iOS and Android, and it also runs on the web at claude.ai. You talk to it like you would text a colleague, typing or speaking, inside a single thread that never resets. Work underway shows up in a Tasks tab. Files produced by bigger jobs land in a Make tab.

Morning brief and check-ins

Every morning Duet goes through your email and calendar, picks out what is important, and can begin on early steps like drafting replies. For the rest of the day it looks at your connected accounts periodically and keeps silent until you are needed. Per the announcement, those looks happen every few hours, since checking nonstop would drain plan limits far more quickly. The help center gives no figure and notes that something urgent can still come in between two checks.

Its own cloud computer and connectors

The agent has a dedicated cloud computer: a virtual machine with a browser, which lets it continue once the app is closed. That browser cannot be watched or taken over by you. Duet reaches your tools through Claude’s connectors and plugins; the help center lists Gmail, Google Calendar, Google Drive, Slack, QuickBooks, HubSpot, Docusign, and Calendly. Microsoft 365 support has not shipped yet. When Duet buys something, it uses Link by Stripe or Shop Pay with single-use virtual cards.

Approvals and safety limits

Before Duet sends a message, pays for something, signs a document, or deletes anything, it asks you. When you are away, it completes everything else and returns with the approval request later. Some actions stay blocked even with your OK: trading securities, transferring money between accounts, and getting into another person’s account. According to Anthropic, your passwords, cookies and app tokens are never visible to Duet: a proxy running outside the agent’s machine adds them to every request.

Model, plans, and known rough edges

The model is Claude Opus 5.5 and cannot be swapped. Duet comes with Claude Pro and Max in the US, through a waitlist, and whatever it does is billed against your plan’s usage allowance exactly like chat messages. Anthropic openly lists the weak spots: Duet can misread a thread, get urgency wrong, or let something slip between check-ins. Personal errands, booking tickets for example, are less mature than work tasks.

Claude Duet vs Grok Bot: how they compare

Grok Bot is xAI’s hosted service for named bots that operate a computer, follow routines, and stop for approval. The core difference in shape: Duet is a single agent, while Grok Bot is a team of them.

The computer. Duet’s machine belongs to that one agent and lives in Anthropic’s cloud. Grok Bot’s docs describe a single cloud computer per account that every bot on the account shares, files and logins included. You can add an optional local computer that runs approved commands on the machine you are sitting at.

How work starts. Duet is proactive out of the box; the morning brief and the check-ins are the whole point. Grok Bot relies more on routines you configure bot by bot.

Price and control. Both are tied to a subscription. Duet is bundled into Claude Pro ($20 monthly) and Max (from $100). Grok Bot is part of paid Cursor plans and of SuperGrok, which is listed at $30 a month. Neither one supports bringing your own model provider, and neither documents a way to run the service yourself.

Krypton: the open source Claude Duet and Grok Bot alternative

Krypton offers this same kind of agent as open source software. It is licensed Apache-2.0 and currently in beta. Run it from the published Docker images or the desktop app, or use Krypton Cloud, also in beta (plans and pricing). Open it in a browser, in the Electron desktop app or in the Expo mobile app.

Chat-first, like Grok Bot. After setup, everything happens in chat. You create bots with names and manage each one inside its own conversation. Every bot keeps its memory, routines, and history. Routines are plain Markdown that can run on a schedule, which means a morning brief is simply a routine you write rather than a built-in feature. You set the boundary where a bot stops for approval, and an audit log records what it does.

Your model. Sign in with Claude Pro or Max, ChatGPT Plus or Pro, or SuperGrok, or add a key for OpenAI, Anthropic, Google, or a local server like Ollama. Bots can each run a different model. If you already pay for Claude, Krypton can put that subscription to work on a computer you operate.

Your computer. Local Docker is the default, with E2B, Daytona, CreateOS, and Boat as optional providers. Team computers are shared between bots, while private computers keep a bot isolated. Connector credentials are stored encrypted on your server, and the API never hands them back. Bots connect to outside apps through connectors, and Krypton Cloud adds a one-click MCP catalog.

Which one should you choose?

Pick Claude Duet if you already pay for Claude Pro or Max, you are in the US, and you want one agent watching your inbox and calendar with zero setup. The trade-off: Anthropic owns the computer and the model choice, and the product is still a preview.

Pick Grok Bot if you are already on a Cursor or SuperGrok plan and want multiple named bots on a computer that xAI hosts.

Pick Krypton if your agents should live on hardware you control, use a model you choose, and you want more than a single bot. What it costs you is the setup: install Docker or the desktop app, sign up, and connect a model. Our guide on how to self-host an AI agent walks through each step, and the open source alternatives hub lines Krypton up against the other agents.

This article compares what each company publishes. It does not test how often any of these agents completes a task correctly. Duet in particular will keep changing while it is in preview, so verify details against the linked sources before relying on them.