A practical guide to setting up Grok Bot, creating specialized agents, connecting plugins, and building routines and triggers that run on their own.
Drafted with Claude from source material, checked by automated verification, and reviewed before release. How we make these.
What is Grok Bot? #
Grok Bot is a desktop and mobile app for running multiple AI agents that stay synced across devices and keep working even when your computer or phone is off. Instead of a single chat window, you get a roster of named agents, each with its own virtual computer, browser session, and set of connected tools. You talk to them individually, they can talk to each other, and each one can be built around a narrow job description rather than acting as a jack-of-all-trades assistant.
The pitch is speed of setup. Building an agent, connecting it to Gmail or Slack, and giving it a recurring routine takes a few minutes and one conversational prompt, compared to the manual node-by-node work required in automation tools like n8n.
TL;DR #
Grok Bot runs a fleet of specialized agents that stay synced across Windows, Mac, and iOS, each with its own persistent computer and browser session.Setup happens through onboarding conversation, not menus: you describe what an agent should do and it asks clarifying questions about which tools and data it needs.** Plugins are shared across agentsonce connected, so authorizing Gmail or GitHub for one bot makes that connection available to every other agent you build.Routines can run on a schedule or on a trigger, such as a new Slack message, letting agents act without you manually starting a chat. Agents delegate to each otherbased on their name, title, and description, so a general assistant can hand off coding work to a dev-focused agent automatically.The teach-a-task feature turns a recorded action into a reusable skill, and agents appear to refine those skills after running them and noticing gaps. Access currently requires a Cursor Ultra plan subscription**, positioning Grok Bot as an extension of the Cursor ecosystem rather than a standalone free tool.
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
How do you install and set up Grok Bot? #
Getting started is a search-and-download process. Searching “Grok Bot” surfaces the official download links for Windows, Mac, and iOS. Installing on all your devices keeps agents and conversations synced, so a routine set up on desktop is visible and controllable from your phone.
Access requires a Cursor Ultra plan subscription. Once installed, the app walks you through creating your first bot: naming it, picking a color and shape, and writing a short description of its job. The interface after that resembles a messaging app, with a sidebar listing each agent like a separate contact or channel, and a settings panel on the right where you can edit an agent’s name, title, description, and view its dedicated computer environment.
How do you create and configure an agent? #
Every agent needs three basic pieces of identity: a name, a title, and a description. The description matters most operationally. Grok Bot uses it the same way it uses skills: when one agent needs help with a task outside its lane, it scans the descriptions of other agents in your fleet and routes the request to whichever one matches. This is what allows an executive-assistant agent to automatically pass a coding task to a separate developer agent without you manually switching chats.
Creating a new agent starts with a short onboarding conversation. The bot asks what you want it around for, ranging from research and writing to day-to-day task management, and then asks what tools or data sources it needs to do that job. If you tell it to help plan your mornings, for example, it will ask where your to-dos and calendar live and then walk you through connecting the relevant plugins.
The general guidance demonstrated in setup is to keep agents narrow. Rather than building one large agent to handle sponsorships, scheduling, and coding, the more effective pattern is several single-purpose agents (a morning planner, a Slack media monitor, a developer) that hand work off to each other. This mirrors the orchestrator-and-subagent pattern common in coding assistants like Claude Code or Codex.
How do plugins and connections work? #
Plugins connect an agent to outside services: Gmail, Google Calendar, Slack, Google Drive, GitHub, ClickUp, and others. Connecting a plugin is a simple authorization flow: click connect, sign in to the service, and authorize access, similar to logging into any third-party app.
The key detail is that plugin connections are shared across your entire fleet. Once GitHub is authorized for one agent, every other agent you create afterward can use that same connection without re-authenticating. This matters if you’re connecting a personal knowledge base or codebase (an “AI operating system” or “second brain”) since every new agent can inherit that context instead of starting from zero.
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
Multiple accounts within the same service are also supported. For Slack specifically, you can authorize more than one workspace or account (a default connection and a separate “work” connection, for instance) so an agent can see channels across different logins rather than being limited to a single workspace.
How do routines and triggers get built? #
A routine is a set of instructions an agent runs automatically, either on a schedule or in response to a trigger. Because routines execute in the cloud, they continue running even if your device is off.
Scheduled routines are built conversationally. Telling an agent to check your calendar and email every weekday morning at a specific time produces a named routine with written instructions (pull calendar events, pull actionable email items, summarize) that you can review and edit directly.
Trigger-based routines fire off events instead of a clock. The trigger options are currently limited to a small set, including new Slack messages, GitHub events, and Teams messages, with the expectation that this list will expand over time. Setting one up involves telling the agent which channel or event to watch and what to do when it fires. For Slack specifically, the connected app needs to be invited into the target channel (using an invite command) before the trigger will actually receive messages from it.
Once live, a Slack trigger works without any manual chat interaction: a message posted in the watched channel is enough to kick off the agent, which then posts a summary or notification back without you prompting it.
What are skills and how do agents improve over time? #
A skill is a recorded, repeatable action. Using a “teach a task” feature, you perform an action yourself inside an agent’s browser (liking posts in a community, for example), and the agent converts that recorded sequence into a skill it can run on its own going forward.
What’s notable is the self-correction behavior described during setup: after running a skill, an agent reviewed its own output, flagged an ambiguity it noticed (not being able to tell clearly whether posts were already liked), and adjusted its own skill instructions to fix that ambiguity without being asked. Agents also appear to accumulate memory and context from repeated conversations, meaning routines and skills tend to get more accurate the more they’re used and refined with feedback.
Is Grok Bot worth setting up? #
For anyone already paying for a Cursor Ultra plan and managing multiple recurring digital tasks (inbox triage, calendar planning, Slack monitoring, community engagement), Grok Bot compresses what would normally be a multi-step automation build into a short conversation. Building a scheduled morning briefing or a Slack-triggered alert took a couple of minutes and one prompt during setup, versus a longer manual build in a tool like n8n. The tradeoff is the Cursor Ultra plan requirement, which makes this a paid add-on to an existing subscription rather than a free standalone product, and the trigger library is currently limited to a handful of integrations. Whether it’s worth adopting depends on how much of your workflow already touches Gmail, Calendar, Slack, GitHub, or similar tools that Grok Bot can plug into directly.
Frequently Asked Questions #
Do I need a paid subscription to use Grok Bot?
Other agents ship a demo. Remy ships an app. #
Real backend. Real database. Real auth. Real plumbing. Remy has it all.
Yes. Trying out Grok Bot currently requires being on the Cursor Ultra plan, after which you can download the app and start building agents.
Can one Grok Bot agent trigger another agent’s help?
Yes. Agents route tasks to each other based on their written descriptions. If an agent’s job doesn’t match a request, it can hand that request off to another agent in your fleet whose description fits better.
What triggers can start a routine besides a schedule?
At the time of setup, trigger options included new Slack messages, GitHub events, and Teams messages, with more integrations expected to be added over time.
Does connecting a plugin for one agent mean I have to reconnect it for every other agent?
No. Plugin connections, such as GitHub, Gmail, or Slack, are shared across your entire fleet of agents once authorized once.
How does the “teach a task” feature work?
You record yourself performing an action inside an agent’s own browser or computer session. Grok Bot converts that recording into a reusable skill the agent can run automatically afterward.