How to Set Up a Bot Squad With One Shared Knowledge Base

Most people use AI assistants the hard way. You open a blank chat, explain who you are, how you like things written, and what rules matter to your work, and then do it all over again tomorrow — or repeat yourself across three different tools. It feels like hiring a helpful assistant who loses their notebook every night.

And once you try to make one single chat handle your morning inbox, your code, your research, and your reminders all at once, the context gets muddled fast.

The fix is to set up a bot squad — a small team of focused assistants where each bot has one clear job, and all of them read from the same shared notebook of Markdown files in a private GitHub repository. Because the files are simple text, your squad never drifts out of sync and you are never locked into one app.

Set up your bot squad in 4 steps

1. Give each bot one clear role

Instead of cramming your entire life into one assistant, split the work across specialists the way a real team works:

• A Team Lead / Chief of Staff that triages your morning inbox, tracks your to-do list, and brings decisions to you
• A Researcher or Writer that pulls facts from trusted sources and prepares first drafts
• A Builder / Coder that works on repositories and opens pull requests for your review
• A Coordinator that watches a shared Slack channel and flags items that need your attention

Write two sentences for each bot describing what it owns and what it must never do without asking you first.

2. Give the squad one shared notebook from a free open-source template

Every bot on your squad needs the same source of truth so you only explain your preferences once. Copy the free, MIT-licensed Agent Knowledge Base Template on GitHub:

• Open the Agent Knowledge Base Template, click Use this template, and choose Create a new repository.
• Set your new repository to Private. The template is public because it holds only blank examples; your copy will hold your real preferences, workflows, and policies.
• Connect your lead bot to the repo and send this prompt: Read AGENT_GUIDE.md and help me build my knowledge base.

You do not have to fill out blank files by hand. AGENT_GUIDE.md tells your bot to interview you one topic at a time, organize your notes into ready-made folders (business, operations, finance, customers, policies, standards, automation, and personal), and keep every new note marked as a draft until you read it over and confirm it. Every other bot on your squad can then read that exact same repository.

3. Let the squad draft, and let a person press send

When you connect everyday apps like email, calendar, or Slack to your squad, keep one golden rule: your bots prepare drafts, and you approve anything before it goes out to another person or spends money. Keep passwords and API keys in encrypted environment variables or a password manager vault, never in chat or in your Markdown notes. For more on safe boundaries, read AI Agents Can Act. A Person Should Still Approve the Work.

4. Put your squad on a schedule and script the repeats

A bot squad really clicks when you do not have to remember to prompt every bot yourself. Add a scheduled routine — like an 8:00 a.m. weekday briefing or a Friday wrap-up — and whenever your bots do the exact same rule-based task more than once, turn that step into a reusable script they all share. See Use a Script for the Repeat Job. Use AI for the Messy One. for how to split the two.

How to connect the same knowledge base across Grok Bot, Cursor, Claude, Codex, and Ollama

Because your knowledge base is just a folder of Markdown files, different members of your bot squad can run on different tools while sharing the same brain.

Grok Bot

Grok Bot is built for running a multi-agent squad on a shared computer. Connect your GitHub account or clone your private knowledge base into your shared workspace folder, then send the prompt Read AGENT_GUIDE.md and help me build my knowledge base. You can spin up specialized teammates (such as a Team Lead, Researcher, Marketer, or Builder) that message each other, share the automation/tool-registry.md list, and run scheduled routines and Slack listeners off the same repository.

Cursor

Use Cursor as the coding specialist on your squad. Clone your private knowledge base repository onto your computer (or add it to a multi-root Cursor workspace alongside your code). In Cursor Settings under Rules, add a project rule in .cursor/rules/knowledge-base.mdc set to Always Apply (or add an AGENTS.md file in the repository root) that tells Cursor to follow AGENT_GUIDE.md and standards/core-principle.md. Whenever you open Agent mode, tag @AGENT_GUIDE.md or the policies folder so Cursor follows your squad's exact standards across every project.

Claude (Claude Desktop, Claude Projects, and Claude Code)

You have three ways to plug Claude into your squad's knowledge base:

• Claude Projects (web and desktop): Connect your private GitHub repository directly to a Claude Project (or upload your Markdown files into Project Knowledge), and paste the contents of AGENT_GUIDE.md into Project instructions so every new chat starts with your context.
• Claude Code (terminal): Add a CLAUDE.md file in the root of your repository that imports @AGENT_GUIDE.md and @standards/core-principle.md, then run claude inside the folder so it loads your rules automatically at the start of every session.
• Claude Desktop local files: Enable the filesystem connector in Claude Desktop settings and grant it access to your local knowledge base folder so Claude can read and draft notes on your computer.

OpenAI Codex (CLI and cloud agent)

OpenAI's Codex CLI and cloud agent automatically look for an AGENTS.md file at the root of your repository before doing any work. Copy AGENT_GUIDE.md to AGENTS.md (or create an AGENTS.md file that points to AGENT_GUIDE.md, standards/core-principle.md, and policies/). Because Cursor and Claude Code also read AGENTS.md, that single root file gives all three coding tools your squad's rules at once.

Ollama (local, offline, and private)

If you want a private member of your squad that runs completely offline on your own computer with zero monthly API cost, install Ollama and pull a fast everyday model such as ollama pull llama3.2. You can connect your knowledge base in two ways:

• Custom assistant with a Modelfile: Create a file named Modelfile in your knowledge base folder with FROM llama3.2 and a SYSTEM block containing your core rules from AGENT_GUIDE.md and personal/example-how-i-like-help.md, then run ollama create my-assistant -f Modelfile and chat with ollama run my-assistant.
• Local web chat (Open WebUI): Open Workspace and then Knowledge, upload your Markdown files as a Knowledge Collection, and attach that collection to your model under Workspace and Models so your offline bot searches your notes before answering.

One shared brain for your whole squad

Models change every few months, and no single tool is best at everything. When your rules live in a private Markdown repository you own, Grok Bot can run your morning routines, Cursor and Codex can ship code, Claude can help draft long writing, and Ollama can run offline — all working as one coordinated bot squad off the same playbook. You can also explore interactive multi-agent workflows live at botsquadai.com or download the starter kit on our Bot Squad™ services page. If you ever get stuck setting yours up, send a note on the contact page.