No-Code Agentist

GROK BOT HOW TO USE Sources: official Grok Bot docs + CLI + your Grok Bot bookmark folder (959 posts) Updated: 2026-09-22 WHAT NEW TODAY - Folder: 959 posts, 4 new versus Sep 21. - Testers: Grok 4.7 is usable from Grok Bot; set Cursor Default Model → Edit → Grok 4.7 with xhigh effort, then hand hard jobs to a Cursor Cloud Agent. An unnamed model inherits that default. - Testers: Grok Bot is now powered by Grok 4.7. - Testers: Cursor coding preserves weekly Bot usage. Model card reports stronger long agent-style coding, engineering, and office work with fewer steps and output tokens. WHAT IT IS - A Bot is a named, persistent teammate, not a one-off chat. - Message it like a coworker; it works in real apps and returns for approval or a human step. - All Bots share one cloud computer with browser, files, and terminal; each Bot has its own screen. - Work continues when the laptop or app is closed. - Grok Bot is the machine that does work; do not confuse it with Digital Optimus or an X Chat Agent. - Official framing: the Bot identifies issues, notifies you when fixed, and continues the project. ACCESS AND INSTALL - SuperGrok and Cursor subscribers have access; listed plans include SuperGrok tiers, Cursor Pro tiers, Ultra, and Teams. - Sign in with the Cursor account that owns the plan and usage. Weekly/free limits reset for Grok Bot users. - Desktop download: Also listed at Desktop supports macOS and Windows; Linux is reported supported. - Mobile: iOS 18+ app “Grok Bot”; Android is available; iPad is reported available. - Link SuperGrok Heavy to Cursor to potentially unlock Cursor Ultra; plans do not stack usage. - Legacy Privacy Mode blocks Grok Bot; choose a supported Cursor data setting. Updates are in Settings → Beta. - Enterprise is available. Grok 4.7 is now reported in Bot; select it through Cursor Default Model → Edit with xhigh effort. - Grok Bot voice is live; voice notes, phone widgets, and account switching are reported. Marketplace and XChat support may roll out gradually. SIGN IN [] Desktop: choose Get started, then Sign In with Cursor in the browser. [] iOS: log in with Cursor, finish in the browser, and return to the app. [] Use the Cursor account that should own usage. [] The first-run tour shapes teammate suggestions; computer setup runs in the background. [] Cmd-D enables voice dictation; remote computer use is easier from a phone. CREATE A BOT [] Choose New in the sidebar or Cmd/Ctrl+N, then Create new agent. [] Edit Profile: name, title, description, and avatar. [] Give one primary job and standing rules; put this task’s instructions in the chat. [] Use focused specialist Bots instead of one catch-all helper; run parallel specialists when serial work is a bottleneck. [] Ask an existing Bot to design a capable Bot when it already has your context. [] Pin important Bots, hide unused Bots, resize the sidebar, and group Bots into sections. [] Duplicate a Bot to reuse its role; duplication does not copy memory or history. [] Delete only when certain; shared files and logins remain. Account limit: 50 Bots and group chats combined. [] The Bot has no in-chat model picker; Cursor Default Model → Edit → Grok 4.7 / xhigh steers Cloud Agents it launches. [] Voice mode supports briefings, exception-only updates, escalation, completion reports, standups, live review, debriefs, and approval gates. [] Marketplace and shareable templates can provide prebuilt context, connectors, and routines; verify availability before relying on previews. HOW TO WRITE A TASK [] State the outcome: what must be finished. [] Name the sources: apps, sites, files, or chats that matter. [] Add constraints: what it must not do and what requires approval. [] Define the deliverable and the review point. [] Scope with Task / Work in / Context / Done means / first N items then stop; meters track agent steps and tokens, not messages. [] Start safely with a cited summary of an attached file; do not change the file. [] Next use one real tool read-only, with a sign-in handoff if needed. [] After a good result, save the lasting format as a skill or routine. [] Use one clear job, allowed sources, a no-guessing rule, and an approval gate before action. [] Calibrate confidence, checkpoint long tasks, show negative examples, gate destructive actions separately, and test edge cases. [] Version prompts and tool configs like code; separate scratchpad from user-facing output; measure cost per successful outcome and review failures. [] Use a plan first, then base/role/current-focus instructions. Include screenshots instead of retyping screens. [] Ask the Bot to draft messages inline for approval before sending. [] Examples: email cleanup, meeting notes, sales prospecting, refunds, podcast digest, website deployment, Starlink-likely flights, and long-talk takeaways. [] Use Grok for research and comparison; use Grok Bot to open apps, change things, send, and finish. [] For hard coding/engineering, set Default Model to Grok 4.7 xhigh and hand work to a Cursor Cloud Agent. AGENT COMPUTER AND LOGINS [] Open Agent Computer from the conversation to watch the desktop. [] Use Network Debugger or route Bot Computer web traffic through your device when needed. [] Take over only for password, passkey, 2FA, CAPTCHA, payment, or other human-only pages; complete the blocked step, then return control. [] Never paste passwords or one-time codes into chat. Use a secure secret request or password manager. [] Browser sessions persist and are shared across all Bots; prefer a connector when one exists. [] 1Password support can pull Mac logins; approve each credential fill. Bot pauses when account authentication is needed. [] Disk Saver can audit caches, temporary files, and duplicates; nothing is deleted without approval. [] Account switching and direct mobile file sharing are supported. [] Connectors can cover Coinbase, X Money, sales/GTM tools, Microsoft Teams, Outlook, Outlook Calendar, OneDrive, and more; keep trades, purchases, and transfers behind approval. [] Keep durable files in /workspace with clear project folders. [] Settings → Beta offers Update Agent Computer (preserve state), Recover, or Reset; Reset can lose recent work. [] The cloud computer is not your Mac or Windows machine. [] Local execution is separate: Settings → General → Agent → Execution on Local Computer. Default policy is Ask every time. [] Give local Mac/iMessage access only on a spare machine. Treat separate Bots as sharing one security boundary. PLUGINS AND CONNECTORS [] Open Settings → Plugins → Add, then authenticate in the browser. [] Type a connector name to attach it; type / to reference a saved skill. [] Installed connectors are account-wide, not isolated to one Bot; multiple accounts can connect to one plugin. [] Follow the Connect card when a Bot asks for a plugin. [] The X path can create a developer account with included credits; some setups may still require a paid API key. [] Image generation is reported live; video editing claims exist, but video generation remains unconfirmed. [] A YouTube link with timestamps can produce a clip in some workflows. [] Microsoft connectors support email, drafts, schedules, meetings, and files. Grok Imagine Image 2.0 can generate images inside a Bot workflow. [] Coding in Cursor can preserve weekly Bot usage; use Cursor or Cloud Agents for heavy coding. SKILLS AND ROUTINES [] A skill is a reusable how-to with steps, rules, output, and approval boundary. [] A routine says when to run the work on a schedule or event. [] Do the task once, make it reliable, save a skill, then automate it. [] Teach a task from Computer view, demo up to 10 minutes, and review the draft skill; teaching records the screen, not microphone audio. [] If teaching is unavailable, ask the Bot to write a skill from completed work. [] Confirm owner, schedule, time zone, input, result, approval, and missing-data behavior. [] Keep event matchers narrow; one channel and one keyword beats every-new-message listeners. [] If the source is unavailable, report failure instead of using old data; test before scheduling. [] A test run does real work. It can send real email; Stop is not a rewind. [] View conversation details → Routines to pause, edit, inspect, or delete. Deleting has no undo. [] A Bot can own 50 routines; recent run records are retained. Deleting a Bot kills its routines. [] Audit forgotten dailies and keep recurring routines off the main agent context when possible. MULTI BOT TEAMS [] Start with one Bot owning an end-to-end outcome; add specialists only when roles are stable. [] Use group chat when the handoff should be visible; Bots can message each other and pass ownership. [] Put sending, buying, deleting, publishing, merging, production, and spend behind a human gate. [] Bots share a computer, so they are not a security boundary. [] A hub or Chief of Staff can route work to specialists; keep one group chat per job or post. [] Give each Bot one job, a supervisor, and an explicit never-without-asking list. [] Use a Red Team Bot to challenge the thesis; no single source closes a claim. [] Show a workflow on screen rather than writing a long spec when possible. [] X Money can send cash or buy merchandise through a Bot; treat every spend as a buy gate. [] Night-shift crews can share one computer, but require a human gate for merge, production, spend, send, and delete. IOS [] Use the same Bots, chats, routines, plugins, and cloud computer as desktop. [] Send text, dictate, attach photos/files, mention Bots, and reply in threads. [] Take over the computer for login or 2FA from the phone. [] Pause or resume routines on iOS; desktop is still needed for schedule editing, history, tests, deletion, and teaching. [] Enable notifications for results, questions, and approvals; approve or deny from the notification. [] Operate or take over the Bot computer from your phone with an internet connection. [] Android and iPad availability may depend on rollout and documentation timing. APPROVALS AND SAFETY [] Put the stop line in the request: “draft, do not send”; ask after showing current versus proposed. [] Desktop supports Allow once, Deny, or a matching rule; iPhone supports Approve once or Deny. [] Auto Review is in Settings → General → Auto-review. Require Approval beats Always Allow. [] Write narrow rules; never allow everything in the browser. [] Do not approve an action you cannot identify. Connect only tools the workflow needs. [] Start read-only; keep spend, send, publish, delete, overwrite, and production behind approval. [] Sign out and revoke connectors when access should end. [] Deleting a Bot does not wipe shared files or browser sessions. [] Check Weekly usage before a large run; one long job can consume trial or weekly credits quickly. [] Use Cursor for coding when it preserves weekly Bot usage. [] Gate destructive actions as a separate confirmation. Test does real work; Stop does not unsend. [] Keep bank accounts and shopping connected only with explicit approval; Bot purchases remain buy-gated. FIRST BOTS WORTH CREATING [] Inbox manager for email and Slack triage. [] Calendar and reservation Bot. [] Research and daily brief Bot. [] Coding Bot that launches Cursor Cloud Agents; set Grok 4.7 xhigh for hard jobs. [] Chief-of-staff hub that routes work to specialists. [] Personal errands Bot for tickets, food, travel, and forms. [] Content Bot that drafts in your voice for approval. [] Accounting, personal, home, shopper, recruiter, CRM scribe, QA, and dashboard-watcher Bots are useful specialist patterns. [] Shopper Bots can research prices, returns, and checkout; keep Link/card purchases behind approval. [] Lecture-to-cheat-sheet, lead-qualifying call, and voice-managed briefing patterns are reported use cases. [] Start in recruitment, e-commerce, real estate, accounting, legal, SaaS, agencies, trades, or founder operations. EVENTS - Official: Grok Bot Galaxy built a company from scratch over three days with engineering, product, and founder sessions. - Challenge: share a Bot that changed how you work for a chance at a Starbase Starship launch invitation. - Galaxy demos showed how SpaceXAI uses Grok Bot to build Grok Bot; a design community shared workflows and feedback. - Build-night roadshow: NYC marketing, LA design, Seattle engineering, San Francisco product, and Chicago recruiting dates were announced. OFFICIAL LINKS - Overview - Get started - Create Bots - Skills and routines - Computer and apps - Approvals and privacy - Mobile - Cursor getting started - SuperGrok Heavy - Launch post - Download - Guides

AI agents like Meta Muse can replace entire categories of point-solution SaaS. I killed my own scheduling app and rethink what’s worth building next. https://t.co/bSAwlg9nWj

Putting five agents in a group chat and hoping they self-organize is how you get duplicated work and nobody checking anybody else. Coordination has to be picked on purpose, and the right pattern depends on what kind of work you are actually running. Generator-verifier is for quality-critical output. One agent generates, another checks the result, and rejected output loops back with feedback until it passes. Reach for this on code generation, fact-checking, and compliance work, where being wrong costs more than being slow. Orchestrator-subagent handles clear task decomposition. One lead agent plans and delegates, subagents handle focused tasks in parallel like security, tests, and style, and all of them return results back to the orchestrator. This fits code reviews, parallel checks, and any task that breaks cleanly into bounded subtasks. Agent teams are built for long-running parallel work. A coordinator assigns tasks through a queue to workers who persist across assignments, accumulating context and domain expertise instead of starting fresh each time. Use this for migrations, multi-step tasks, and anything that benefits from an agent actually specializing. Message bus fits event-driven pipelines. Agents publish and subscribe to events through a shared bus, so new agents can join without anyone rewiring the system. This is the right call for security operations, growing ecosystems, and workflows that change shape over time. Shared state works when the goal is collaborative findings. There is no coordinator at all, agents read and write directly to a shared store, and findings become instantly visible to every other agent. Reach for this in research synthesis, where you want no single point of failure and real-time collaboration between agents. Bookmark this before you default to one coordination pattern for every kind of agent work.

Meet Jev: an AI that can't write a single sentence, and that's the whole point 😄 It just makes decisions, fast and cheap. Big claims (193x faster!), some serious fine print. I broke it down 👇 https://t.co/ajdhlJx8Sz

Zuck said, “You bring the API.” So I built and submitted a Muse connector for WeOutsideDC. The docs were sparse, so I wrote a walkthrough with build, test, and submission prompts you can paste into your agent. https://t.co/dfzm9dFMtJ