Former Tesla AI director Andrej Karpathy shared a method for building a personal knowledge base with LLMs this April: put your data in a folder and let AI write and maintain a wiki on its own. On October 18, X user Argona (@Argona0x) shared a version of this approach adapted for xAI’s always-on AI agent, Grok Bot: put three Bots, each with its own memory, in the same group conversation, assigning them to collect information, write the wiki, and find errors. Here’s a summary of the approach, along with related features and limitations from the official Grok Bot documentation. This is a workflow designed by a user, not an official xAI feature.
Three bots share the work; a checker hunts for errors every night
Argona’s setup has three bots each handling a different role in the same group chat:
Reader: Takes in every link, PDF, and voice message you send, without ever modifying the original files.
Librarian: Whenever new information arrives, rewrites every affected wiki page. The author says this usually means updating 10 to 15 pages at a time.
Checker: Runs automatically every night to find contradictory content, outdated claims, and orphan pages that aren’t linked to by any other page.
The workflow has three rules. When you ask a question, the answer is archived as a new page, so the knowledge base grows through use, not just through reading new material. When a new source conflicts with an existing page, the checker flags it for you to decide rather than silently overwriting it. Any page that hasn’t been corroborated by a new source for more than 30 days is marked “unconfirmed.”
The author stresses that a second brain can also “remember things incorrectly with great confidence,” so every page must state three things: the claim itself, its source, and the date it was last verified. He sums up the three roles in one sentence: the reader collects, the librarian writes, and the checker questions.
Why use Grok Bot: it keeps organizing things even after you close your laptop
In another long post on October 13, Argona explained that most popular second-brain setups online run on your own laptop, so the organizing stops when you shut it down. Each Grok Bot, by contrast, has its own computer in the cloud.
According to Grok Bot’s official documentation, all bots share the files, browser state, and logged-in apps on the same cloud computer. Closing the app, laptop, or phone won’t interrupt background tasks or routines; bots can also share context in a group chat. The documentation also advises asking bots to check current sources when making important decisions, rather than relying on memory alone. This is also why the author recommends storing knowledge in files rather than in chat memory: raw data goes in the raw/ folder, organized pages go in the wiki/ folder, and any bots added later can access them.
Routines have usage limits. The official documentation states that routines must be scheduled at least 5 minutes apart, with a maximum of 50 routines per bot. Argona recommends scheduling only a few runs per day, saying that a routine running once an hour could use up a week’s worth of bot usage on your plan in a single day. For Grok Bot plans and pricing, see the complete Grok Bot guide.
The first prompt to paste into Grok Bot
Here is Argona’s suggested starter prompt translated into English. You can paste it directly into Grok Bot:
Help me build a second brain using three bots in the same group chat. Keep my original data untouched in raw/, maintain wiki/ yourself, archive useful answers as new pages, and check for contradictions every night. Show me the folder structure first, and don’t create anything until I approve it.
He recommends starting with a topic you’re actively working on, rather than throwing your whole life into it from the start.
In his October 13 post, he also outlined seven steps: first, paste in 20 emails you’ve sent so the bot can create a profile about you and capture your writing style; then set up folders so new material is automatically archived, generate a summary page every morning, and save recurring tasks as skills. Finally, record each correction in a rules file so the same mistake isn’t repeated. This sequence aligns with the official documentation: do it manually first, save it as a skill once it’s stable, and only then hand it over to an automated schedule.
This article, “Grok Bot Second Brain Tutorial: Three Bots Organize Knowledge, Find Errors Automatically Every Night,” first appeared at .
【Further Analysis】
【My Further Perspective】
This technological breakthrough reflects the shift in AI agents from “single-question-and-answer tools” to “always-on autonomous workflows.” By dividing tasks among agents—collecting, writing, and reviewing—users are simulating a human knowledge-management team and addressing the shortcomings of LLMs that rely on chat memory, are prone to forgetting, and lack the ability to work in the background. This means personal productivity tools are evolving toward fully automated operation with no human intervention, significantly reducing the maintenance costs of knowledge management.
【Market Impact】
The growing adoption of cloud-based AI agents and automated workflows will directly benefit major U.S. cloud computing companies, such as Microsoft, Amazon, and Google, as well as AI infrastructure providers, by driving increased demand for computing power. At the same time, stocks associated with Elon Musk, who is behind xAI, or cryptocurrencies such as Dogecoin (if it is integrated into a payment ecosystem in the future) may receive a short-term boost from market sentiment. In the software-as-a-service (SaaS) sector, startups focused on knowledge management and automation may face structural pressure on their valuations if they cannot adapt to the impact of native AI agents like these.
【What to Watch Next】
If xAI later opens up more comprehensive developer APIs or enterprise-level access, it is expected to spark a wave of software development focused on “AI agents working together.” Investors should closely monitor whether cloud service providers can effectively convert capital expenditure into revenue, and whether automated agents for financial trading and earnings analysis emerge in the market. These will be key indicators for gauging the commercialization of AI.
【Risk Reminder】
When AI agents operate autonomously in the cloud without rigorous logic checks, they may accumulate false knowledge through “hallucinations” or consume excessive computing resources in automated scheduling, leading to high costs. In addition, entrusting large amounts of personal or corporate confidential data to cloud-based bots may pose serious cybersecurity and privacy risks.