用 Grok Bot 搭三代理自检第二大脑

Kodzilla @k0dzilla原推发布于 2026.10.29

在 X 上打开原推

I TURNED GROK INTO A SECOND BRAIN THAT CHECKS ITSELF. Imagine dropping a PDF, a link, or a voice note into a chat and never having to explain that information to AI again. The setup uses 3 specialized agents: 1. READER Collects links, PDFs, and voice notes. Saves the original material without changing it, so nothing gets lost or distorted. 2. LIBRARIAN Turns incoming information into organized wiki pages. Updates related topics, connects new findings to existing knowledge, and saves useful answers for later. 3. CHECKER Runs a nightly review. Looks for contradictions, outdated claims, missing sources, and pages that haven’t been verified recently. Here’s where it gets interesting: → Ask a question, and a useful answer can become a new wiki page. → Add a source that contradicts an older claim, and the system flags the conflict instead of silently rewriting history. → Leave a page without fresh evidence for 30 days, and it gets marked as unconfirmed. Every knowledge page keeps three things: • What it claims • Where the information came from • When it was last verified Want to try the same setup? Paste this into Grok Bot: “Design a second-brain system with three agents in one group chat: Reader, Librarian, and Checker. Reader must preserve all original sources in raw/. Librarian must maintain wiki/ pages, connect related information, and save useful answers as new pages. Checker must review the knowledge base nightly for contradictions, stale claims, and missing references. Every page must include its claim, source, and last verification date. Show me the proposed structure first. Do not create anything until I approve it.” Start with one subject you actually work on. Give your AI a memory it can inspect, update, and question.