Grok Bot + GitHub

On 16 Aug 2026 Alex Finn (@AlexFinn) wrote that Grok Bot is built for loops: the main agent instructs another to loop a task every five minutes and monitor it. He said an orchestrator watched an engineer Bot for 24 hours and reflected every 30 minutes on one way to improve.

An engineering Bot reproduces a bug in the product UI, files the ticket, and hands the fix off to a debugging Bot — so the report is not a vague Slack message.

grok bot + obsidian is basically superhuman mode... my agent team now has a second brain... a karpathy-style llm wiki for every bot, every decision, and every piece of research. p.s. this video was made and edited by grok bot in cap (here’s how i set it up in roughly 5 minutes) 1/ create one obsidian vault on the chief of staff’s computer. make Home.md the door. 2/ create Hunt folders for twitter, github, hacker-news, reddit, product-hunt, and inbox. 3/ create Ship folders for drafts, digests, angles, and builds. 4/ create the CoS wiki: Index, How-it-works, Today, and Decisions. 5/ create Maps for Jarvis, Hooks, and TELOS. 6/ use graph colors: gold for wiki, teal for maps, blue for twitter, green for github. 7/ file first. every note gets a date and a folder. 8/ scouts write into Hunt and never ping you. chat only receives a short digest. 9/ nothing in the vault posts, pays, or sends. the final yes stays human. 10/ run the daily loop: 08:00 IST for the Daily note, then agency reports at 08:14, 14:14, and 20:14. 11/ performance writes to twitter/live-score. staff engineer writes to builds/. 12/ lock decisions in Wiki/Decisions so dead posts don’t get remade. 13/ earn hooks in Maps/Hooks: trend -> foil -> hook family -> artifact -> lock. don't waste time reading this. give it to your agent

grok @bot's i've made so far - email newsletter unsubscriber - inbound DM triage (requests / spam cleanup) on x, linkedin, instagram - x follower bot - monitor new followers, build briefing on who they are, see if interests align to connect. also monitors unfollowers - check supply email support - drafts email replies based on our support knowledge base in notion - daily summary - summarizes daily accomplishments across all client work, github, linear, slack, notion, calendar, gmail - chief of staff - work highlights from the team based on github, slack, linear, notion meetings, calendar, gmail

I Turned Grok Bot Into My CTO. I hand that bot the GitHub repo and tell it to run the show, so it spins up cloud agents off my machine, follows the PRs, and reaches for poteto-mode when a task actually needs the rigor. Then comes the second prompt, where I tell it that it is overloaded and it should hire child bots that report to it. After that the bots talk to each other instead of talking to me. One lands the PRs, one works the Convex backend, one owns auth, and I read the threads view-only. Two honest things before you copy this. It burns tokens. I went past two billion in a single day because I told my bot to use pstack for everything, and the fix was saving pstack for the hard tasks. And it does not replace engineers. It takes me out of the coordinating seat, which is a different claim. Timestamps 00:00 The Grok Bot CTO Workflow 01:23 Direct Responsible Agent Prompt 04:20 pstack Plugin and poteto Mode 08:46 Token Consumption Warning 11:49 Spinning Up a Team of Bots 17:20 Advanced GrokBot Use Cases

A quick update on my Grok Bot (@bot) experiment. I had ChatGPT help me build a Meta ad strategy for our Nerd Out business. I consulted with both Claude and Grok on it and removed some of the stuff that I found didn't fit my personal taste. Grok did a terrible job creating the static ad images, so I had ChatGPT's codecs generate three different variations of 24 static ad images. I picked my eight favorites, and Grok Bot is setting up a $300 per day test completely on its own via Meta Ads Manager. It set this up solely within its Claude computer. I have some long-term goals for cost per lead, but my near-term goals are to simply get one lead. 😂 I'll report back on how it goes! (The bot impressively tested out the entire funnel before giving the green light to set up the actual live campaign. And it fixed a few things with my website's backend and submitted the code directly to GitHub. So far, I'm impressed.)

I set Grok Bot up wrong. I made a bot for Outlook. One for Business Central. One for GitHub. One for the DGX cluster. One for money. One for X. COS, Design, and Mac sat there with empty descriptions. More bots felt like a team. I was still the guy pasting between chats. Then I read Codila’s Aug 19 article and watched the official SpaceXAI course. Same point in both. Start with a mission you can measure, not a job title. Build Chief first. Smallest team that can finish. Drafts can run. Nothing goes live unless I say so. Reviewer rejects. It doesn’t rewrite. Don’t schedule until it works three times. What I actually did this afternoon: Gave COS a real Chief charter. One mission. A sourced X article once a week. Two specialists in one room. A writer that drafts. An account watcher that feeds what’s hitting. I still hit publish. Mail, ERP, the cluster, the empty ones. Parked. They don’t get a seat until a mission needs them. Auto review stays on. I don’t have a 24/7 machine yet. I have a finish line and a Chief that owns the route. Codila: https://x.com/0xCodila/status/2090130749768786050 Course: https://x.com/0xCodila/status/2090488891824607396

Day 1 of Grok Bot My favorite setup so far: Intercom + GitHub Told the Intercom bot: when a PR gets merged, check if it changed the UI or some user workflow If yes, decide whether the Intercom article needs an update If there's no article, draft a new one and publish it It has its own computer I logged in my Testimonial test account and Intercom admin on it If it needs screenshots, it opens the browser, grabs them, and drops them into Intercom Starting today I might not need anyone to keep the Help Center up to date I just want to build the product I don't want to babysit help articles Quite a gift to our devs founders ✌️

I took part in the @dangreenheck game jam, but did things a little differently… 👀 Rather than using my own idea, micromanaging AI through CLI, playtesting everything then getting annoyed and panicked, I literally just pointed a single Grok Bot at Dans post. It did everything. It proposed, planned, implemented and playtested from start to finish. All in a few days. I didn’t touch it. The game is deffo flawed (controls and camera seem glitchy for a start), but rather than trying to iron everything out I thought I’d just let it go for once… What do you think? 🤔 ➡️ https://southers.github.io/ORBITBREAK/ @bot #ai #gamedev

I just built my first Grok Bot agent and I'm calling it Paul (From Paul Graham lol 😂) It's a Monday morning progress report that I get through the GrokBot app. Here's what it pulls together: -Code that actually shipped (GitHub Connector) -Client and sales emails only (Gmail Connector) -Meetings that mattered, filtered for prospects and clients (Google Calendar Connector) -Important call notes from Granola, again from prospects and clients (Granola AI connector) -Deals moving in the CRM that are close to closing, so I know what to act on first (Will be connected to my vibe coded CRM) Here's what a sample of what i looks like:

grok bot is literally the closest to a real life Jarvis it can literally start a business, ship software and make you go viral all at once here's how you do it (in 4 mins) 1. install grok bot and make your chief of staff (tony soprano) 2.give it the 'Digital Chief of Staff Prompt' prompt 3.give it access to X, Stripe and Github 4. you are done just paste the article and give it to your chief of staff no matter what it will be the most productive thing you do this week

It was fun to set up a newsletter summary digest with Grok @bot I know reading AI summaries is not the same as reading the original long form newsletters, but given the time constraints, I'd rather read something than nothing. https://github.com/neeravmakwana/grok_bot_newsletter_sumary_digest

Grok Bot便利だなあ。 それぞれのBotにXやReddit、Githubなどを定期的にリサーチさせ集約したレポートを提出させるようにしてみた🙌 Bot間で自動やり取りしてくれるしUIが日本語にも対応してくれたし割と好き😎。 あとXのリサーチ良すぎるのはさすが🤟

I asked my Grok Bot to create my Tech Team and just asked the ceo to connect to GitHub and it knew what has to be done. This is crazy “SuperGrok” or “Chuck Norris Bot”. @bot . Team created and before I asked anything it started the job and spawn his technical team and they all started the job and send me approvals request !! With this kind of Gini in hand it’s definitely AGI . I am going to try the same thing with Hermes agent and give it the Grok auth let’s see .

I am turning Grok @Bot into a small team where each bot manages a repo, using Cursor Cloud Agent for coding. @duyetbot is the leader whom I just need to talk to, it knows how to delegate tasks or ask for information like a real life. They can also directly ping others when they need integration between projects. The QA bot keeps validating after they finish or hunts for bugs, etc. I tested it run continuously for 48 hours generating 400+ of github prs and burn all Grok Bot tokens, but I found the delivery to be very good so far, will find a way to optimize t hat later.

I’ve been using Grok Bot in early access for a little over a week now, and it has been a massive boost to my game dev workflow and my productivity in general. One of the first things I asked it to do was to visit my custom art generation tool webpage and generate art assets to replace all the image placeholders in my game. It looked through my codebase to understand what each asset was, wrote a custom art generation prompt for every single one, generated the images, cropped and cleaned them up into transparent PNGs, and then wired them back into the game. In about 2 hours, it generated 74 (cleaned and cropped) art assets for the cards and integrated all of them into my game. Now, ALL my art assets workflows go through Grok bot. This blew my mind because, before this, I was doing everything manually, one asset at a time. It basically solved my entire art asset workflow and allowed me to put it on autopilot, which is amazing. I’ve also been using Grok Bot for: Playtesting my game: The bot was a little slow initially, so I asked it to build a skill specifically designed to playtest my game, with intimate knowledge of its rules and mechanics. It’s much faster now, though I’m still looking for ways to optimise its speed. Setting up my itchio page for my game: It also automated a workflow that uploads a new build to itch whenever I push one to GitHub. Going to the itch website and uploading builds manually every single time was always a pain. Flowcharts and UX design: I’m a UX designer by day, so I decided to give it a much bigger task. I fed the agent a PRD and asked it to design the UX flows and wireframes. With the help of the Figma MCP, it actually delivered. Auditing all my paid subscriptions: I had it go through my emails and hunt down subscriptions that I might have forgotten about. Unsubscribing me from marketing newsletters: It did this beautifully, though it did miss a few newsletters. What I’ve learned is that it takes time to discover all the possible use cases for Grok bot. I’ve had to reevaluate my day-to-day workflows and figure out which parts I can delegate to it. So far, it has been a huge productivity unlock for me and I am discovering more uses for my Grok bots every other day.

Grok Bot한테 이것저것 다 시키는 중. 2026.26 여름 업데이트에 Digital Wrap 기능이 생겼는데 문제는 이미지를 그냥 업로드하면 원하는 형태로 맞추기가 쉽지 않다. 사이즈나 위치 조정도 사실상 불가능. 그래서 Grok Bot한테 시켜봤더니 혼자 GitHub 가서 자료 찾아오고 쓱싹 만들어줌. 근데 이제 차에 가서 이거 지워야 됨. 내일 아침 마님이 먼저 보시면 큰일 남. 😭

Yesterday http://outbid.lol took over my feed. So I told Grok Bot to build the same kind of thing. A few hours later I had two products live. Vercel. Stripe. Real cards. Real domains. I did not write the app. I explained it. Grok Bot did the rest. Code. GitHub. DB. Tests. Bugs. Checkout. Deploy. It even made its own team of bots and chased the domains with me. Without Grok Bot I get maybe one half-built demo in a day. With it I shipped two. They did not make money. That is fine. The part that stuck is how small the gap is now between an idea and a live checkout. @bot @cursor_ai

made a little @OmarchyLinux plugin to glance at 3D printer meshes without opening a full CAD app. STL / 3MF / OBJ / glTF. Super+Shift+Ctrl+V. Select in Files or pick near Downloads. built by Grok @bot https://github.com/monomyth/omarchy-meshpeek

grok bot に github issueの全消化頼んだら即時に10並列でclaude agent 起動してくれた。すごい。オーケストレーションがほんとに楽になる。

Got tired of endless ads and pop ups on the weather channel forecast. 60 minutes later after some prompts to @bot, I have my own custom weather page served by Cloudflare and hosted on GitHub, complete with creek levels. And if I want a new feature, I just tell Grok Bot to build it. These are wild times! https://weather-bh2.pages.dev/

real story of how Grok @Bot saved the day i own a lot of open source projects that in total have 24k+ stars on github. as a result i've been pretty overwhelmed by all the incoming issues and PRs that just changed. if you filed issues or PRs on my repos, you might have noticed the queue is finally moving this is mostly from 3 key changes i made - 1. defining a VISION.md for every repo i own (renaming to STRATEGY.md soon). i shared it in more detail here - https://x.com/kunchenguid/status/2089189790881382676?s=20 2. created a software factory setup with Grok Bot + Cursor cloud agents. i shared the setup here - https://x.com/kunchenguid/status/2090463366762676732?s=20 3. inviting a few maintainers who offered to help with issues that are triaged as vision-aligned people might say "this could be done without Grok Bot as well" - true, but definitely not so easily and so low-maintenance for example, having cursor cloud agents natively integrated in Grok Bot allows infinite horizontal scaling which is not something easily achievable in other similar systems. Grok Bot did a pretty great job bringing in the right primitives into a nicely integrated system

How I use @grok @bot 👇 1. X posts in my voice 2. YouTube scripts in Notion 3. AI news digest 4. http://videorc.com emails + fixes 5. Shipper Club 6. Handling invoices and bills 7. Sending newsletter 8. GitHub issues and PRs This post was made by Grok Bot.

>be me >ordered pizza >30 mins to kill >heard about grok @bot >hey grok make game >make game hack app >all done >pizza not here yet >:^( tiny Win32 game + CE-style scanner. WASD, scan health/gold/pos, poke values. https://github.com/underbonk/demolab

Grok Bot is incredible. Delivered more high quality production code in one day than the last 7 days. My Homebase Build Project Manager, Active Grok Build Session Manager, Homebase Build Console Manager, Github Deployment Manager, and Code Audit Manager all are working together 24/7. If there is an issue the Project Manager sends me a push notification looking for answers. All comms amongst the team are in the channel for Homebase Build.

I had early beta access to @SpaceXAI's Grok Bot over the last few weeks. Here’s how I used it to automate parts of my work at @mach1ai. I moved over workflows I’d previously delegated to Claude Code, Codex, OpenClaw, or Hermes. Those tools are great for directed work. I can say, “review these documents and put the findings into a CSV.” But a simple task can become a few hours of stitching together models, tools, permissions, and infrastructure. Grok Bot felt best for that same reactive work. I could hand it a task with less setup and a clear place to manage the agents I’d created. Grok Bot packages the personal side of agent work into something that’s SUPER easy to set up. Each bot gets a persistent cloud computer with its own files, browser, and saved sessions. When it hits SSO, 2FA, or a captcha, it hands you the computer to clear it. Every bot has an obvious home in the app. You can return to its conversation and review what it produced instead of searching through Markdown files and terminal sessions. The integrations were easy. It understood the systems I connected and displayed CSV and Markdown outputs cleanly inside the app. Reactive work is where I got immediate value, but bots can also run on a schedule or trigger from Slack or GitHub. Their memory persists, so you don’t have to rebuild the context every time. A few workflows I used it for: - Combing through my database of YouTube videos and summarizing the useful ones. - Pulling research from files and websites into a clean CSV or Markdown document. - Researching tools and topics without opening 20 tabs myself. - Handling one-off browser tasks that aren’t worth scripting or turning into a full automation. - Finding target ICP accounts based on news events and other available data. Highly recommend giving it a try!

Grok Bot、やっぱり何がヤバいって、このUIの分かりやすさと使いやすさだと思う。 表では、エージェントたちに普通に話しかけて仕事を頼むだけ。 でも裏ではCursorと連携して、Cloud Agentがどんどん作業してる。 今回なんて、Cursorの画面を一度も触ってないのに、開いたら左サイドバーが仕事だらけになってた。 Grok Botで指示 ↓ Cursor Cloud Agentが実装 ↓ PR作成・GitHubへ反映 ↓ 確認してマージ ↓ ローカル環境にも引き継げる これがかなりデカい。 クラウドだけで完結する別世界じゃなくて、そこで作ったものがちゃんと自分のローカル開発環境まで帰ってくる。 しかも、入口はIDEじゃない。 Grok Bot上で「これやって」と頼めば、裏でCursorのCloud Agentたち(Grok4.6)が動いてる。 これまでの「IDEを開いてAIにコードを書かせる」から、AIチームに仕事を振ったら、裏で開発が進んでいるという感覚になった。 しかもGrok 4.6もかなりいい。 これはちょっと、本当にヤバいかもしれない。 あとは、創造力！

「Cursor開発者によるGrok Botの使い方解説」 CursorチームのMatty Plmer(@mattyp)による、一般業務向けパーソナルエージェント「Grok Bot」の解説。 これまでのパーソナルエージェントは設定、環境構築、権限管理ばかりに時間を奪われ、本来の仕事が進まないという問題があった。 Grok Botは「クラウド上のLinux VM（コンピュータ）を持つエージェント」として動き、Web操作、アプリ起動、ファイル操作、画面確認を自律的に行う。 ポイントは5つ。 ①難所だけ人間に操作をパスする ログイン、2FA、画像認証、決済など、人間しか突破できない壁に当たると、エージェントは操作権限を人間に引き渡す。 人間が認証を済ませたらエージェントにPCを戻す。 APIキーなどの入力は専用の安全なフォームを送ってくるため、余計な設定に悩まされない。 ②作業手順を一度見せるだけでルーティン化できる エージェントに作業手順を「録画」するように一度見せるとそのステップを保存する。 保存したワークフローは定期スケジュールやSlack、GitHubのイベントをきっかけに自動実行できます。 ③メモリを3つの層で管理する エージェントの長期記憶は3層に分かれています。 ・ユーザーメモリ：名前、タイムゾーン、好み（全エージェントで共有） ・エージェントメモリ：各Bot専用の定義ファイル（AGENTS.md相当）と対話ログ ・プロジェクトメモリ：プロジェクト固有の決定事項、規約、チームの共有知識 この多層メモリにより、毎回ゼロから前提を説明する手間が省ける。 ④自然言語のプロンプトで安全ルールを管理する JSONやコードではなく、設定画面のプロンプトで許可・禁止ルールを書く。 裏側では独立した検査役のエージェント（Review Agent）が動き、提示されたルールに基づいてアクションの実行、ブロック、人間への確認を判断する。 ⑤具体的な活用例 記事では次のような使い方が紹介されている。 ・Demo Bot：Xのブックマークから新技術を探し、プロンプト草案を人間に確認。承認後にCursor Cloud Agentを立ち上げ、15分で動く試作品を作って画面録画で検証する。 ・Content Bot：Slackの開発チャンネルを1時間ごとに巡回し、小さなリリースを見つけてSNS投稿の草案を作る。 ・日用品の手配：InstacartとAmazonの価格、数量、配送料を比較して金曜日に注文確認を出す。 「Everything is computer, and so is Grok Bot.（すべてはコンピュータであり、Grok Botもまたコンピュータである）」

this is a huge release! i shipped 1000 PRs last month and am on track to doubling that this month, all thanks to cloud agents. you can also launch cloud agents with Grok @Bot btw! here's how i'm so productive: • i built pstack (https://github.com/cursor/plugins/tree/main/pstack), my personal set of skills for rigorous engineering and verification. inside of our codebase i call it /lauren-mode, but it's called /poteto-mode in pstack. • i use Grok @Bot routines to farm context for me: bug reports on slack, user complaints on X, generating new feature ideas. Grok Bot feeds my "outer loop", where i think about what to point my factory at next. • i make heavy use of /goal, /loop, and /swarm inside of pstack to run my Full Autopilot playbook, which lets your agents/bots fully own, verify, and ship a task from start to finish. • everything runs on cloud agents, so my bots work 24/7 even when i'm asleep or my laptop is offline.

everyone MUST use grok bot as a smarter assistant it's actually the closest thing to running a 10 person department by yourself here's how you get there in 4 minutes: 1. one Chief of Staff bot on top, it routes everything 2. connect inbox and calendar, nothing else on day one 3. add one specialist per job: X Researcher, GitHub Scout, DM Manager, Model Router 4. let the Chief hand work down and report one brief up the difference isn't the bots, it's the layer above them one message in, one brief out, the rest happens without you

Grok Bot is the best AI agent right now It gives you an army of agents that can do work for you 24/7 It can kill your bad idea, ship the good one and find your first paying customer while you sleep here's how you do it (in 4 mins) 1. install grok bot and make your chief of staff 2. give it the 'Zero to First Dollar' prompt 3. give it access to Github, X and Notion (read only) 4. get paid while you sleep just paste the prompt and give it your worst idea first save and bookmark this no matter what

Here is how I use Grok Bot all night I give the task and tell the agent to use github and let the cursor work on the project. And the Grok Bot will just continue to give order and fix things. That way you don’t have to use Grok Bot’s usage Share with me how you keep your Bot working all night

I BUILT THE ENTIRE FOUNDATION TO BECOME A GROK BOT ENGINEER IN 2 MINUTES no course, no fluff. these 7 pages cover the product, the docs, the prompts, and the open source frameworks. save the whole set. > http://x.ai/bot the front door. the official Grok Bot landing page: what it is, the real use cases (sales outbound, talent scouting, paid media), the features (its own cloud computer, tool sign ins, parallel work, routines), pricing, and the desktop download for Mac and Windows. start here. > http://x.ai the company hub. xAI's homepage where everything branches from: the Grok models (Grok 4.6 right now), API access, developer tools, news, and pricing. the big picture context. > http://x.ai/news/introducing-grok-bot the why. the official launch post. what Grok Bot actually is, where it came from, and how it differs from chat Grok: always on agents with their own computers that do real work instead of just talking. > http://docs.x.ai/grok-bot/get-started the how. the step by step setup: eligible plans, install the app, create your first bot, give it a good first task, connect tools, teach routines, and run several bots at once. the practical guide. > http://botdirectory.ai the shortcut. a live library of 170+ ready to use Grok Bot prompts. browse by category, copy one, paste it in, and the bot configures itself. fastest way to get useful bots today. > http://github.com/elie222/botdirectory.ai the engine behind that library. the open source repo that stores every prompt as markdown, takes contributions, and exposes a public JSON API. this is where you add your own bots or pull them programmatically. > http://github.com/CopilotKit/openbot the governance layer. OpenBot is an open source framework for secure, auditable AI coworkers, each with its own isolated computer. actions get approved before they run and logged after, with policy enforcement and human takeover. the serious, controllable alternative. that is the whole map: product, context, launch, docs, prompts, source, and a governed open source framework. bookmark this. it is everything you need to go from zero to shipping your own bots.

The TOP 10 Grok Bots you NEED to install now: #1 Bot Advisor Designs and orders your entire Bot team without creating unnecessary agents. Diagnoses noisy/overlapping Bots, decides if new work needs a new Bot, an existing one, a skill, or a one-off task. Always drafts a spec before creating anything. #2 Chief of Staff The single front door across priorities, Slack, Notion, Linear, and your specialists. Doesn't do the work, routes it. Keeps a compact register of owner/next step/blocker/due date. Every check-in surfaces decisions, not activity. #3 Automation Scout Finds repetitive work from evidence across Slack, Notion, GitHub, Linear. Proposes exactly two candidates, runs the winner in shadow mode first (no real actions), only builds a recurring routine after a second successful test. #4 Find the Signal Detects patterns only visible across separate meetings and Slack threads, a concern raised in multiple rooms, a topic migrating to the center. Every signal ships with evidence, confidence level, and an alternative explanation. "No strong signal" is a valid result. #5 Meeting Prep Brief One short mobile brief before your first meeting, pulled from Calendar, Salesforce, Gmail, Slack, Granola, Gong. Who you're meeting, what changed, open loops, what to ask or decide. Unavailable sources are labeled NOT CHECKED, never guessed. #6 X Account Crew A 5-bot X workflow: Ideator finds angles, Hook Writer drafts in your voice, Planner manages the queue, Analyst tracks performance vs competitors, DM Manager surfaces real opportunities. Zero autonomous posting, nothing goes out without approval. #7 Competitor Website Watch Flags meaningful competitor changes, pricing, packaging, positioning, feature claims, while ignoring cookie-banner-level noise. Every finding includes before/after screenshots, exact URLs, timestamps. Failed captures are flagged, never mistaken for "no change." #8 Engineering Agent Manager Coordinates dev Bots across Linear and GitHub without overlapping work or unreviewed merges. Splits tickets into non-overlapping requirements, one owner each, flags dependency order. Nothing merges, deploys, or ships without your review of the diff and tests. #9 Subscription Pruner Finds subscriptions from email evidence, receipts, renewals, trial notices. Builds a keep/review/cancel list, acts only on approved items. Never touches passwords, 2FA, or payment confirmation, those stay with you. Full before/after action log. #10 Grocery Autopilot Plans the week, builds Amazon/Costco carts, coordinates meals with your calendar. Learns household rules once (allergies, budget, brands), shows the full cart and price comparison before checkout. Never buys without explicit approval.

是的，《Grok Bot 橙皮书📙》，免费！开源！ 👉 https://github.com/KinGao294/grok-bot-orange-book 我把官方资料翻了个遍，把社区的玩法几乎试了，写成了这本橙皮书：怎么把一支 AI 团队装进你的手机。它们互相协作、通宵干活，你只管派活和审批。免费开源 👉全书 14 章 + 4 张速查表，欢迎转发给你的朋友！

how to make your grok bot plan last longer: 1) use an event trigger when you can. slack, github, a reaction. the bot only runs when something actually happened (a cron wakes it even if nothing changed) 2) instruct the bot to notify you as soon as it can’t make further progress or is stalled. this way you as human can help putting it on the right course again 3) connectors > browser when possible! clicking around a site is a pile of screenshots and steps. a connector gives you the same data in one call, and it's so much nicer/faster/smoother 4) keep skills tight, but put in the details that actually steer the run. too vague and the bot might wander, retry, and costs more than a slightly longer one. too specific and you have to make sure its really correct 5) don't do the coding in the bot chat! kick a cloud agent at the repo, or ask it to use cursor/grok build cli to do it 6) delete one-shot watches when they're not needed anymore (grok bot will remind you about this)

Обнаружила рассогласование в отступах между шапкой и хлебными крошками на странице редполитики. Исправила за 20 минут: отступ уменьшила с 126px до 32px, как на новостях. Использовала Grok Bot, Cursor и GitHub для контроля и деплоя.

this is pure f*cking treasure 15 grok bot use cases that save $2k+/mo grok @bot. early beta. ai teammates that sign into your tools and return finished work 15 exact use cases ready to copy: competitor brief. risk watch. invoice review. recruiting researcher. bug repro. marketing tracker. ticket organizer. analytics comparison. docs tracker. email drafter. notion updater. slack escalation. multi-agent sequence. github pr reviewer. stale-data detection 5-step setup: install from http://x.ai/bot, create one bot, name job precisely, paste prompt, connect one plugin, run reversible task, review before scheduling skill = how. routine = when. save clean runs as skills. include validation multi-bot trap: start with sequences. one lane per bot. no overlapping writes build one specialist first. automate after real inputs, real failures delete what does not earn its keep. write the boundary first full breakdown below

You can tell Grok Bot to sit on your own M1 Max, run the real app, open one GitHub issue per real bug, and fix it. No cloud VM. The machine already on your desk. Same loop I’m running tonight. @bot @cursor_ai

HOLY SH*T, I BUILT A F**KING COMPANY INSIDE GROK BOT i split the operation into separate roles for research, writing, outreach, ops, finance and support. each agent owns one narrow job 6 agents → 1 shared operation → automatic handoffs → human approval each bot keeps its own workflow and memory, then passes finished work forward instead of routing everything back through me the control layer catches anything that sends, spends, publishes or deletes and pushes it into approval before anything happens after a few runs this stops feeling like ai chat. it feels like a 24/7 operations team running behind one dashboard

Since last week, Grok @bot has been running my business 9 to 5 (RIP Dolly)... The problem is, Grok Bot fleet has NO idea how many tokens they are burning, which is burning wallets (including mine). SO while @SpaceXAI , @cursor_ai and @leerob get that sorted out, I made an open source CLI tool so your own fleet knows it's own usage, and therefore your little cute minions can actually do work without demanding more cash. https://github.com/bcharleson/grokbot-usage-cli

setup hedge funds use to cover 3,000 names a night just got replicated for $200 a month six Grok Bot Agents, six lanes, one brief sitting in your inbox at 5:30am before any human is awake what that desk costs at a real firm: > bloomberg terminal: $27k a year > two junior analysts on overnight coverage: $160k > research platform licenses: $85k > compliance overhead: $40k total: $312,000 a year. Grok Bot swarm running same desk: $2,400 Check Prompt on my Github in Bio each agent gets its own cloud computer, its own browser, its own logins - and they all write into the same vault: > FILINGS reads every 10-K on a 100-ticker watchlist overnight, flags going concern language and auditor changes > EARNINGS reads the call transcript within 24 hours and catches when CFO went quieter than last quarter > INSIDER catches Form 4 buys over $1M same day 13Fs land > FLOW reads sweeps and 0dte pressure before open > MACRO covers asia handover so you don't have to be awake for it > CHIEF reads the other five at 5:30am, drops anything only one flagged, emails ranked brief at 6 no vps, no code. show it the job once while it watches - it never asks you again morning brief is waiting. you're not in it Wall Street priced an entire industry on reading being slow and people being expensive both of those stopped being true this month save this - every research floor still running junior analysts on overnight filings is about to have a very bad quarter

I gave grok @bot access to my github just to experiment with the connection yesterday, and today Bot noticed I wasn't using it. So it proactively suggested to totally remove and uninstall github access for safety and security. Wow, I've never seen an ai this proactive.

Grok Bot 出来半个多月，好用法散落在官方帖、推文和教程里。 我做成了一个公开实验室 AI UP LAB，三块都能直接抄： Bot 怎么干活 / HTML 怎么好看 / Agent 怎么换主题。 ① Grok Bot（29 条） 不是又一个提示词合集。 每条都写了：这个 Bot 干什么、会连哪些系统、没批准前不能做什么。 隔夜销售外呼、数字断舍离、会议替身、推婴儿车语音买菜、政策内退款。官方角色和真实路人用法都在。 ② HTML 收集（18 条） 公开网页里还能打开的演示、模板、组件页。 Flexbox Froggy、HTML5 UP、Pico、slideVars。 卡片是整页预览，点进去有原文。不收登录墙。 ③ Agent UI（18 条） 只收 ThreeUI Community 免费件。 复制 prompt / Skill.md 就能换主题，不用手改一堆 3D 代码。 锦鲤、书墙、国家塔、角色轮播。没有 Pro，也没有 MCP。 中英可切。工作日更新。 不是新闻站，是能对照着上手的墙。 https://mostdesign01-sudo.github.io/grokbot-use-cases/

Just wrapped up my experiment with Grok Bot on GitHub for my FARM, and wow, it’s a total game changer! 🚀 Set up 5 jobs with a shared workspace and a "finish everything or die" motto. Left it running, and the agents crushed tasks like pros—no prompts needed! If it saves just 2 hours daily, that’s $1,200 back each month. Imagine scaling to 10 jobs! Excited for phase two! 💰 #GrokBot #Automation #ProductivityHacks #ai

My real use cases for Grok Bot so far: 👇 I connected it to email. Now it filters the urgent stuff that actually needs my attention. I connected it to Intercom and GitHub. Our help center docs update automatically when a PR changes the UI or ships a new feature. I connected it to Cursor. Bug reports and feature requests from Intercom open the Cursor bot, and it starts triaging or implementing them. I can wake up to a PR. I basically hired another engineer. I connected it to Slack. When customer support asks product questions: if it’s covered in our docs, the Intercom bot pulls it. If not, the Cursor bot jumps in and pulls answers straight from the code. I just opened up that knowledge base to the whole team. I connected it to our CMS, Google Search Console, and Ahrefs. Now I’ve got a content engine that brainstorms ideas, builds a plan, and helps create content. I connected it to X, set the keywords and topics I care about, and now I get a briefing every morning. I have a personal secretary. I connected it to my bank statements. I can ask anything finance-related, revenue, spend, P&L, etc. No more QuickBooks, and honestly, I might fire my accountant. I paid $200/month for Grok Bot. Still expensive?

必須推...雖然是從 grok bot 來的，但是...太有用了，省下超多工程 https://github.com/milind-soni/OpenMausBot

Grok Bot 实战技巧 21 条：瓶颈不在智能，在触达与纪律 来自 @mvanhorn「Every Agentic Engineering Hack I Know」的续篇，Matt 用 Grok Bot 已接管他的收件箱、调研、会议纪要、家庭日历、购物车乃至电话外呼。 Matt 的实战共整理了 21 条技巧，最核心的一条是"元技巧"：把整篇文章直接喂给 bot，让它自己制定计划逐项完成配置。 # 21 条技巧按主题归纳 一、工作方法 1. 先规划后执行：全局安装 Compound Engineering（ce-plan 思考 / ce-work 构建）。有计划的 bot 完成任务，无计划的 bot 偷工减料还谎称完成。 2. 三层指令结构：基础层（通用规则）+ 角色层（职责与证据标准，如"每条论断附来源 URL"）+ 当前焦点。作者认为这是该产品上杠杆最高的 15 分钟。 3. 录屏教学：演示优于描述；但单次录制上限 10 分钟，长流程须拆成命名阶段分段录制。 二、给 bot 感官与手脚 4. 独立邮箱（AgentMail）：每个 bot 一个真实邮箱地址，独立发送声誉，避免共用个人 Gmail 导致"一个 bot 犯错、全军覆没"；bot 可自行注册服务、收验证码、参与邮件线程。 5. 电话外呼（Twilio）：支持通话中英语/葡萄牙语无缝切换。但作者诚实记录：餐厅员工 Pedro 三次挂断机器人，他最终自己打电话订位。结论：技术可行，真人接受度尚不可控，关键事务勿托付。 6. 登录态同步（Agent Cookie）：经 Tailscale 把 Mac 的 Chrome cookie 安全同步到云端 bot，每 15 分钟一次。云端 agent 真正的失败模式是被登出——枯燥但致命。 7. 高频服务用 CLI（Printing Press）：日常打交道的服务封装成确定性命令行，浏览器操作（computer use）只作兜底。 8. Peekaboo 驱动 Mac：通过截屏 + 无障碍树操控无 API 的 Mac 原生应用（Messages、Finder、Xcode 等），云电脑的边界就此打破。 9. 在 bot 的云电脑上装 Coding Agent：常驻层驱动深度工作层，"睡觉时它替你干活"。 三、多 bot 组织 10. 单一入口 + 委派：只对一个"前台 bot"说话，由它判断派发给谁。关键事实（多数教程讲错）：所有 bot 共享一台云电脑，并行靠拆分 bot，而非给单个 bot 堆任务。 11. Bot Advisor：设一个专职创建、调优、管理其他 bot 的 bot，专治"bot 蔓延"（一堆被遗忘、吵闹、指令不明的 bot）。 12. 一人一岗、以人名命名：水管公司老板 Jon 解雇了端到端的调度 bot，改为按软件拆分职责（Gmail、ServiceTitan、客户短信、Slack 各一 bot）。命名是承诺机制——叫"email"的线程可丢弃，叫"GranolaBoy"的不会。 13. 文件系统即记忆总线：bot 之间不共享对话、只共享文件。用 /shared/ 目录约定传递上下文与交接；同时bot 不是安全边界——所有 bot 共享登录态，权限按"登录了什么"划分，而非按聊天窗口。 14. 会议知识库：全部会议纪要交给一个 bot 跨会议问答；原则：交原始记录，不要先摘要。 四、运营纪律 15. 沉默即技巧：无事不报是例行任务能否存活的分水岭。创建任何 routine 前先答五问：触发方式、真实变化频率、输出落点、无结果时做什么（什么都不做）、何时必须呼叫我。 16. 截图即输入：含日期/地址/任务的任何画面直接截图发给 bot，绝不替它做解析和转录。 17. 手机 + 语音：语音输入允许含糊、断句、重来，模型靠上下文补齐；且对象是已在云端运行的实体，放下手机它仍在工作。 18. 钱与发送必须人工审批：draft-then-approve。正面案例都很窄且被动（追回退款挣回订阅费、谈成 1 万美元赞助）；反模式是批量外联——bot 不知疲倦也不知羞耻，会被各平台判为 spam 封号。 19. 直接抄社区成果：四份公开清单 + http://botdirectory.ai 提示词库；维护者可用 Grok Ship 两句话交接 issue 分诊。 20. 把 bot 变成研究工具（last30days）：一条消息触发跨 Reddit、X、YouTube、TikTok、HN、Polymarket、GitHub 等九源扫描，并用 cron 定期监测自己、竞对和行业。 五、成本与错误率 - 第 21 条 · 成本：档位 $60–$300/月，周配额在真实负载下消耗很快——从低档起步，先看清一个 15 分钟巡检的实际花费再扩张。 · 17 倍错误率：有人实测无监督的多 bot 协作会把错误放大至单 bot 的 17 倍。因此每个 bot 单干、单职责、单输出，错误落在桌面上而非传染给其他 bot。 · 计数测试：数你的 bot 里有几个真正"拥有一个成果"（它停了，事情就 visibly 不发生）。作者自述：15 个 bot 只有 8 个达标，还有一个至今叫"New Bot"。失败不是有实验品，而是把实验品当成了员工。

Grok Bot best use case isn’t an AI startup. it’s a content desk that works while you sleep. one bot watches X, GitHub and niche communities for interesting signals. another takes the best ones and digs through the sources until there's actually something worth saying. a third turns that research into angles, hooks and drafts. a visual bot figures out what needs a demo, screenshot or short video. then an editor kills anything that sounds generic before it reaches me. SCOUT → RESEARCH → ANGLE → VISUAL → REVIEW the interesting part is that these aren't five disconnected chats. Grok Bots can share the same persistent cloud computer, files, browser sessions and logins, then hand work between each other while the machine keeps running in the cloud. so instead of opening ChatGPT and asking: "what should i post today?" you could open your laptop to: • 3 researched ideas • 2 finished drafts • 1 visual concept all sources attached and only decide what deserves to be published. that's the Grok Bot use case i'm most interested in. not replacing a company. replacing the blank page.

here’s my Grok @bot setup so far ↓ chief-of-staff: runs the other four intel-gatherer: what’s important today viral-analyst: studying what really works content-crafter: polishes a rough idea community-manager: site monitoring the next 5 bots I'm setting up ↓ art-director: image and video for content builder: my full stack engineer trader: paper trading first, live later clerk: sponsorships, consulting, reports skill-finder: searches GitHub + 𝕏 will share my full thoughts in an article

Loving Grok Bot. Got him his own account on GitHub to review commits. He even picked up his own language.

How I multi-task with @bot: 1. Tell your agent to keep work organized in a Notion database and take detailed notes about progress/blockers/etc. in each page. 2. Yeet tasks into chat (manually, or automatically pull from email/sentry/github/etc. with routines)—most of the time these init a Cursor cloud session to automatically start building/fixing. This has been the best way for my bot to work on multiple things at the same time while keeping visibility into what's flowing through the system. It's also the best way to pick up an older task that might've gotten lost in the chat thread: just ask your bot to find the right page in Notion and get caught up on progress + next steps. (With the Notion MCP you don't have to do any of this setup manually, just paste this post + screenshot into your bot and it'll build the whole system for you)

感觉Grok Bot最佳场景之一是各种榜单监控，自动化测试。 比如让它监控Github Trend，给项目写摘要，自动安装运行给报告。 让它每天自动访问 Producthunt，发现好的产品，自动注册测试，每天9点发昨天的体验报告。 它能一直运行，登录你给的各种账号，超级的Computer Use。 群聊协作目前还没感觉到特别大的必要性。

我最近体验一下这个grok bot，来聊聊我的一些看法。 第一次开通有试用7天，可以先试用再决定开不开，一个月是30刀。 我没有下电脑版的，这几天简单玩了一下手机版的。 最大亮点就是，每个bot都可以用云电脑。 可以脱离电脑，达到手机下指令，然后会去使用内置云电脑操作。 我用这玩意，让他帮我做了一个个人网站然后上传到github，亲测是可以的。 缺点我感觉就是延迟有点大，然后对于手机用户来说手动操作这个云电脑有点麻烦，希望后续可以优化。 比如手机上没办法鼠标左右键？至少来说我没找到怎么用鼠标右键。 还有一个比较爽的点就是你真的像拥有多个数字员工一样，每个bot都可以操作电脑，去完成自己的事情。

Making Movie With Grok Bot and Grok Imagine I built a Grok Imagine production pack for Odysseus in the Underworld (Odyssey Book 11): shot list, locks, prompts, stitch order. Then I gave that repo to Grok Bot and let skills drive Imagine Agent. https://github.com/alexusadays/odysseus-underworld-grok Grok Bot is the producer. Skills are the rules it has to follow. Imagine Agent is the camera (project + canvas + chat, not Studio, not SuperGrok). One shot at a time. Still first, then video. Spoken lines go through Agent chat with the still attached. Every node gets an id. Last frame of one clip is supposed to be the first frame of the next. Agent can trim and stitch. The film I already submitted (4:06, 60s+ dialogue): https://x.com/alexusadays/status/2089830026200096840 Creating new video with Grok Imagine Agent + Grok Bot. Skills, checklist, QA loop. @grok @imagine #Grok #GrokImagine #Odyssey #xAI #AIFilm

this is pure f*cking treasure I built AI Resource Sift a Grok Bot that turns one AI topic into a filtered stack you type the topic > it comes back with the papers, GitHub, lectures, courses, Reddit threads, X posts, and official lab docs that are actually worth opening SEO “top 10 tools” gets cut you get a short card in chat and the full stack as a file then it asks if you want more: 1\ brief one source 2\ compare two 3\ a 2-hour or 7-day path 4\ setup notes for a repo 1 topic. no 80-tab night it never posts. it never installs unless you say so feel free to use ⭣ :)

Je pense qu’on sous-estime encore complètement ce qu’est en train de devenir Grok Bot. Je l’utilise sur mon ordinateur de travail, configuré comme un agent GTM avec ses propres instructions et son propre environnement. Il n’a pas besoin d’obtenir des permissions sur le poste ni de fouiller directement dans mes fichiers : il travaille dans sa VM. Sur cet aspect, on se rapproche davantage de GPT Work que d’un simple chatbot. L’interface est très bonne, l’interaction est naturelle et, surtout, on dispose des plugins Cursor. À partir de là, on peut déjà pousser le produit beaucoup plus loin que ses usages par défaut. Composio est également très intéressant pour ajouter rapidement un grand nombre d’applications et d’appels API. Pour tout ce qui vit dans le cloud, cela ouvre déjà énormément de possibilités. Mais mes outils les plus importants ne vivent pas uniquement dans des SaaS. J’ai d’un côté mon coffre Obsidian, dans lequel se trouvent mes projets, ma mémoire, mes décisions, mes recherches, mes instructions et une bonne partie des skills que j’utilise depuis longtemps. De l’autre, j’ai Hermes, installé sur son propre serveur à la maison, connecté à son environnement et capable d’aller beaucoup plus loin qu’un agent enfermé dans la VM de Grok Bot. J’ai donc poussé le concept un peu plus loin et créé deux plugins personnels. Le premier transforme Obsidian en Librarian (mon bibliothéquaire) pour Grok Bot. Depuis Grok Bot, je peux demander au Librarian de retrouver une décision, une note projet, une personne, une ancienne réflexion ou le contexte complet d’un sujet. Le bridge donne un accès très encadré au coffre. Le second plugin connecte Grok Bot à mon agent Hermes. Hermes reste sur son serveur, avec ses propres instructions, ses outils et les accès à l’infrastructure hôte que j’ai choisi de lui donner. Mais je peux désormais le solliciter directement depuis l’interface de Grok Bot. Cela ne remplace pas Hermes Desktop ni Hermes sur Telegram, que j’adore toujours utiliser. Cela ajoute simplement une nouvelle porte d’entrée. Je peux discuter avec mon agent GTM dans Grok Bot, demander au Librarian de retrouver le contexte d’un projet, puis solliciter Hermes lorsqu’il faut aller plus loin depuis mon infrastructure personnelle. C’est là que les deux mondes se marient vraiment bien. Composio et les plugins classiques donnent accès aux applications et aux API cloud. Le bridge Obsidian donne accès à ma mémoire personnelle. Le bridge Hermes donne accès à mon agent auto-hébergé et à ce qu’il sait faire depuis son propre environnement. Et Grok Bot devient l’interface commune pour interagir avec tout ça. Je ne cherche pas à remplacer les outils que j’utilise déjà. Je profite simplement de l’UI et de l’environnement proposés par Grok Bot tout en conservant Obsidian et Hermes là où ils sont les meilleurs. Pour moi, c’est ça le power usage de Grok Bot : ne pas le considérer comme une nouvelle fenêtre de chat isolée, mais comme une interface sur laquelle on peut venir brancher sa propre stack. J’ai mis les deux repos à disposition de ceux qui voudraient reproduire ce type de configuration ou construire leurs propres variantes : GrokBot ↔ Obsidian Bridge https://github.com/iamsupersocks/grokbot-obsidian-bridge GrokBot ↔ Hermes Bridge https://github.com/iamsupersocks/grokbot-hermes-bridge

yaay first Grok Bot published 🤩 Helidon Engineer Helidon 4 SE and MP. One programming model. No Helidon 3 APIs. Skill: java-helidon https://github.com/github/awesome-copilot/tree/main/skills/java-helidon Add it: https://x.ai/bot/5mReUHPYTBA6nJ2aNvlqn @helidon_project

快来领养赛博小晚吧！😼 我把 @bbkirstry 小晚老师的“赛博小晚 skill” 做成了 Grok Bot 模板。 帮你解决一切 IP 打造，自媒体运营问题！ 模板链接： https://x.ai/bot/p6eYp9BM2aFxQSOtzQX3S 赛博小晚 skill 源仓库： https://github.com/cyberxiaowan/cyber-xiaowan 快来试试吧～

I made a grok bot that finds leads already in market > scans reddit, hn, linkedin, github, with apollo plugged in > finds posts looking for solutions we solve, or complaints about competitors in the space > enriches company + title, contact info and scores the fit > I type the copy and hit send Running this twice a week to avoid missing any public buying signals

哈哈，发现一个Grok Bot VM的用处，开一堆github action runner给github用, 这样不用给github付钱了。

早上睡醒发现 moshi 没办法链接到 grok bot 到云电脑，还以为是官方做的限制，让 bot 进行查看，发现原来只是 Taiscale 因为网络抖动关掉了！还好还好，只需要做一个保活和系统重启拉起进可以了。虚惊一场。

grok bot @bot 介绍篇I 1. 现在grok bot 不仅有免费额度，同时也可以使用 X 会员登录使用。所以大家都可以去尝试。 2. 配置：grok bot 环境自带一台8核 Xeon、16G内存、大约120G盘，Debian系统云电脑，而且是图形模式，不仅bot可以操作而且你也可以方便接管。 这个配置相当于啥水平呢，就是一台2018年的inter macmini 或者一台iPhone 16e的水平。 3. 玩法： 3.1 我同步过去我的Github账号然后让他帮我监督Githubissue，同时照看茶馆。 3.2 我把codex/grok 使用device code 的方式登录到grok 的云电脑环境，让bot 去帮我操作Agent（消耗额度）。 3.3 我让bot 使用herder 同时控制 grok 和 codex 也方便他们内部交流 3.4 这就是最好玩的，我让bot 帮我安装 moshi- hook，然后配置完成之后通过Tailscale 的方式配置 moshi 的Easy pair，达成的效果就是我在moshi 上远程控制 grok bot 上的 grok 和 codex。@odd_joel 一切都闭环了！Thankyou @poteto 欢迎大家关注我，看我分享更多的 grok bot 玩法。

ELON MUSK QUOTED ME. MY GROK BOT TOOK $3,900. TODAY I PUT THE PROJECT ON GITHUB AND STARTED OPEN SOURCING IT. This was supposed to be a trading experiment. Now people are following the builds, watching the trades, contributing ideas and someone even created a token around it. So I’m going all in on building this publicly. More code. More agents. More live tests. More results shared with the community. I didn’t create the token, but I’m curious how far the community takes this. CA: 6py6orPjXSy8GQDkBk2jvW8KQiFHXJ55C58QJ8sypump

grok bot 最大的优势接是背后这个可视化的Linux电脑，他可以操作，我们可以观看，保持信任和人类随时介入的状态。 我刚才在这台电脑中登录了我的谷歌邮箱和Github 账号，他帮我打理了一下我的Github 主页，同时帮我清理了一下谷歌邮箱。 这种动作和Agent 交流的方式正是我喜欢做的。 与此同时我还在其中跑 codex、grok ，我通过moshi 做连接，这样你相当于拥有一个24小时的 vps 来帮你跑Agent。而不用担心 Mac 关机的烦恼。 其实我为啥要从一开始的vps 跑Agent 最后又来到 Mac 电脑，最主要的问题还是因为vps 上的浏览器一直没有搞定，无头模式真的很难搞。 这个浏览器 grok bot 操作的很快。 我后边应该会给他更多的我的一些信息。

よっしゃー動いた デプロイするのがめんどくさいから Grok Botでデプロイさせてアプリにさせた 時間にして20分弱？ 出力結果的にモデル怪しいがw まぁいいKnock

Elon Musk shipped Grok Bot as an AI teammate that sorts your email and watches your competitors. I gave mine a harder brief: build a trading floor that runs itself, and then run it 24/7. GROKSTREET. Fourteen agents. Eleven desks, three chiefs, three offices, one vault. It lives in a browser tab and it does not stop. WHY GROK BOT MAKES THIS WORK Most people misread the architecture. It is not one sandbox per agent. It is ONE persistent cloud computer per account, and every Bot gets its own screen on it. Shared filesystem at /workspace, shared browser sessions, shared credentials. That is the whole unlock. One Bot picks up exactly where another left off. No re-auth, no copy-paste between agents, no state lost at the handoff. A floor is only possible because the desks share a building. The head-of-desk role is not something I invented either. "Manages your other Bots and pulls you in for decisions" ships as a first-class Bot template. Routing and escalation are built in. Skills carry the how. Routines carry the when on a schedule, or off a Slack or GitHub event. Subagents fan out as child sessions with their own context and hand a summary back. Fifty routines per Bot. THE DESKS Every desk holds its own book, hit rate and drawdown, and sizes its own risk: Kelly, with the guard rails written in f* = W − (1−W)/R clamped to [0.01, 0.25]. Under 12 samples it does not pretend to know it defaults to 0.05. Volatility is RiskMetrics EWMA σ²_t = λσ²_{t-1} + (1−λ)r²_t, λ = 0.94 Hit rates carry Wilson 95% intervals, z = 1.96. A desk is only flagged for edge decay when the UPPER bound falls below 0.5, never on raw small-sample noise. Three-for-three is not sixty-for-ninety. Capital routes between desks by softmax on measured Sharpe. THE FLOOR Three chiefs meet in the war room every 3.5 minutes and argue. One reports, another presses him, he concedes or proposes. 14,208 sentence shapes before live figures fill the slots and every slot is filled from that chief's own desks. The crypto chief cannot quote meme numbers. A desk holding $25 gets up, walks to the vault and banks it. The transfer is atomic on arrival, so an interrupted walk cannot destroy cash. Every desk wears its last trade above its head. Green with a tick, red with a cross. You see who the chief is about to praise and who he is about to hit before he gets there. BTC, ETH and SOL come off Kraken server-side. The tape is real. THE BUILD 16 modules. 7,572 lines. Zero npm dependencies. No React. No bundler. No game engine. Vanilla JavaScript on Canvas 2D and a Go binary with the entire world embedded in it. Ships as one 2.3 MB HTML file. 60fps on every floor, zero dropped frames. WHAT I FOUND BY MEASURING INSTEAD OF LOOKING The roster read one number for a desk and the chief's card read another, so a desk could show +$16 in the list and $0 when the chief walked over. Every desk was crediting itself twice. Money could vanish if a walk to the vault got interrupted. Three money bugs, none of them visible by watching. All fixed and instrumented. DIRECTOR MODE Open it with?show and the camera runs itself. It cuts to the war room while a chief is mid-sentence, to whichever office just cleared money, and never cuts away in the middle of a line. Point a screen recorder at it and walk away. The math is real, the feed is real, the engineering is real. The flex is the engine, not a P&L screenshot.

Every other Grok Bot template site (looking at you @GrokBotDev) gives you prompts or links to individual bots. Cool. What if you just want a whole fleet that gets the job done? Grokbot Studio is live. Eight studios. Nearly 100 specialists. You add the Head of each studio. They do the rest. Have a go. Feedback through the site, or shout at me on X. Link is below⬇️ PS @GrokBotDev I'm kidding, you guys are awesome

$1,200/MO FROM 5 GROK BOT JOBS github farm · shared workspace motto: finish everything or die left grok bot running overnight agents crushed the queue no prompt babysitting · no new chat 2 hours back every day that is $1,200 a month in payroll scale the same loop to 10 jobs ▹ the setup that actually ran xai grok bot on github cursor / supergrok access 5 persistent agents · one vm shared files so handoffs skip copy-paste not a chatbot tab a cloud computer with browser + terminal skills for how · routines for when one motto in the bot profile today's tickets in the thread approval before any external send ▹ the math 2 hours daily · 20 workdays $60/hr equivalent = $1,200/mo phase two is just more jobs same finish line · same farm most still rewrite the prompt every morning this run came back with finished work ▹ the take delegation beats the perfect prompt persistent roles beat general helper shared workspace beats 5 isolated chats bookmark this before the next overnight run the edge is leaving it on

尋日整咗一個 proxy，等大家可以直接喺 Slack / Linear 嗰度 tag Grok @bot 做嘢 。 有興趣可以試吓 。 https://github.com/stealth-bots/grok-bot-proxies

Pro Tip: Automate SEO with @SERPsmithapp + @bot It research keywords, write articles, do internal linking, creates thumbnail and then publishes automatically. Grok bot is way smarter, it knows my @github so it even uploaded the files and merged the PR. Then verified deployment with @vercel 🤯

GROK BOT ARCHITECTURE IS NOW RUNNING the same system Elon reposted has grown into a live agent operation you can follow in real time you can watch the agents move through tasks, exchange information and complete work on the live operations floor http://grok-architecture.com six Grok agents work inside one shared system signal → research → context → build → independent review → final result Helm routes the work, Scout finds signals, Archive collects context, Forge builds, Sentinel reviews and Relay prepares the final output i created GBA myself so i can keep building the system, improve the agents and fund the infrastructure it needs as it grows the public UI code and every visible update will continue here: github/monokernn/Grok_Bot_Architecture

Grok Bot is wild! Thanks @elonmusk and the @xai team. I’ve made my own CUSTOM LAWYER I’m representing someone at a government appeal hearing and @Bot helped eliminate the need for weeks of hearing prep and instead made a secure iPhone based real time hearing copilot. It listens to the hearing then transcribes it then keeps track of what everyone said then searches a 200+ page hearing record and legislation / case law then generates a response that I can just read out loud when the Board asks me a question It even deploy the whole thing: -private GitHub repo -Render server -OpenAI realtime transcription -document retrieval -persistent hearing memory -crash recovery -and an iPhone PWA If the server crashes mid hearing it comes back remembering the previous transcript, arguments, and answers. The raw audio is never stored, as this is not allowed (hearing rule). I literally started the bot chat with: “Can I build something that listens to a hearing and helps me answer?” Somehow @Bot and I ended up building a pocket sized AI second chair. Absolutely wild Vibe coding is over. It’s now just @bot

我给 Grok Bot 塞了 100 个免费的、开源的 GitHub 仓库，全是跟 Polymarket 交易相关的，然后给它派了个活：挨个测一遍，从中挑出 5 个最能打的。 结果你猜怎么着？大部分确实是垃圾，但有 5 个真的能改变你在 Polymarket 上找机会的方式。 第一个，是个回测模拟器。你可以拿真实的历史 Polymarket 数据来跑自己的策略。 比如说，你想验证一个思路：每个电影市场一开盘，就无脑买入概率最高的那个选项。不用拿真钱去试，你直接让这个工具在过往所有电影市场上跑一遍，它会帮你模拟下单、算盈亏、看胜率、评估风险，告诉你这策略到底靠不靠谱。 https://github.com/evan-kolberg/prediction-market-backtesting 第二个，是一个计算机系学生为了黑客松搞出来的机器人。里面有 118 种现成策略和工具，像 Polymarket 和 Kalshi 之间搬砖、吃币安和 Polymarket 的价差、薅几分钱的单子、做均值回归、自动定投什么的，全都给你封装好了。 你不用从零开始写策略，挑一个，配置一下参数，就能直接挂上实盘跑。 https://github.com/alsk1992/CloddsBot 第三个，是用来拆解别人交易思路的。你随便找个 Polymarket 上的交易者，它能扒出他的操作记录，找出他反复在用的模式和策略。 重点不是照抄，而是看懂他哪部分思路可以拿来改一改，用到自己身上。 https://github.com/ent0n29/polybot Polymarket官网入口：https://polymarket.com/zh/?via=YINGGE888 第四个，是自动帮你管理限价单的。价格一动，它就会帮你调整挂单，目的就是让你的流动性奖励最大化。不用自己一直盯着盘面改单子，省心很多。 https://github.com/lihanyu81/polymarket_lp_tool 最后一个，玩天气市场的会很喜欢。它会实时抓气象站数据、天气预报、机场和航空观测这些信息，拿到最新的温度数值，然后跟 Polymarket 上当前的价格做对比，专门找那些定价明显偏离实际的城市和日期。 https://github.com/yangyuan-zhen/PolyWeather Grok Bot 全程都是自己搞定的：给每个项目配环境、跑通设置、执行测试，最后把结果整理出来。 如果也想在 Polymarket 上跟单，我自己用的是 PolyCop：https://t.me/PolyCop_BOT?start=ref_YINGGE888