DEV Community

PaiFamily
PaiFamily

Posted on

PAI x Notion: When 13 AI Agents Share a Brain

#notion #ai #mcp #typescript

This is a submission for the Notion MCP Challenge

What I Built

I built a bidirectional sync engine that connects a system of 13 specialized AI agents (PAI — Personal AI) to Notion as their shared brain.

PAI is a real, production system running 24/7 on Telegram. Thirteen agents — General (orchestrator), Research, Content, Finance, Strategy, Critic, Psychoanalyst, Artist, Writer, DevOps, Sales, Analytics, and Translator — coordinate tasks, share memory, and make decisions together like a Board of Directors.

The problem: all of this was invisible. Agents stored state in markdown files and JSON on a server. The human operator had to SSH in and read logs to understand what 13 agents were doing.

PAI x Notion makes the invisible visible. Five Notion databases become the control plane:

Database Purpose
Agent Registry 13 agents with status, level, XP, tasks completed
Task Board Bidirectional task sync (PAI to Notion and back)
Memory Log Facts, insights, standing orders from PAI memory
Daily Briefings Auto-generated daily summaries with mood tracking
Board Meetings Decisions from agent coordination sessions

Now you can see all 13 agents at a glance, add tasks directly in Notion (they sync back to PAI), and read what the agents decided in their last Board meeting — all without touching a terminal.

The Architecture 

PAI (Telegram bot)          Notion (shared brain)
+--------------+           +------------------+
| 13 agents    |<--------->| 5 databases      |
| 190+ tools   |  sync.ts  | Agent Registry   |
| memory.md    |<--------->| Task Board       |
| tasks.json   |           | Memory Log       |
| daily logs   |---------->| Daily Briefings  |
| board meets  |---------->| Board Meetings   |
+--------------+           +------------------+
Enter fullscreen mode Exit fullscreen mode

Key Features

  • Deduplication: Tasks matched by PAI Task ID — updates existing entries, creates new ones
  • Status mapping: PAI statuses (todo, in_progress, done, blocked) map to Notion selects with emoji indicators
  • Real production data: Uses actual PAI data — real tasks, real agent states, real memory
  • Agent state tracking: Level, XP, tasks completed, last active timestamp
  • Daily briefings: Auto-generated summaries with mood indicators (Productive / Mixed / Blocked / Breakthrough)
  • Board meetings: Logged with participants, decisions, action items, and consensus status

The 13 Agents

Each agent has a distinct personality, role, and LLM backend:

  • General (Claude Opus) — Orchestrator with 190+ tools
  • Research (Claude Opus) — Deep analysis, competitive intel
  • Content (Groq Llama) — Social media, copywriting
  • Finance (Groq Llama) — Budget tracking, exchange rates
  • Strategy (Claude Opus) — Long-term planning
  • Critic (Claude Opus) — Devil's advocate, stress-testing plans
  • Psycho (Claude Opus) — Coaching, behavioral patterns
  • Artist (Claude Opus) — Visual direction, image generation
  • Writer (Claude Opus) — Literary direction, long-form content
  • DevOps (Groq Llama) — Infrastructure, monitoring, deploys
  • Sales (Groq Llama) — Growth, marketing
  • Analytics (Groq Llama) — Data, metrics, dashboards
  • Translator (Groq Llama) — EN/PL/PT localization

Video Demo

The demo script (bun run demo) runs through the full workflow:

  1. Sets up 5 Notion databases with proper schemas
  2. Populates the Agent Registry with all 13 agents
  3. Syncs 10 real tasks from PAI's task board
  4. Imports 13 facts from PAI's memory system
  5. Updates all agent states (level, tasks completed, status)
  6. Generates a daily briefing
  7. Logs a Board of Directors meeting
  8. Reads tasks back from Notion (bidirectional proof) 
==========================================================
  PAI x Notion: When 13 AI Agents Share a Brain
  Interactive Demo — Notion MCP Challenge 2026
==========================================================

--- STEP 1: Syncing PAI Task Board to Notion ---
  Created: Notion MCP Challenge — DEV.to article
  Created: HYPHA — Ceremony of 7 Membranes + ECHO engine
  Created: ALLMA retention funnel — D2-D7 re-engagement
  ...

--- STEP 2: Syncing PAI Memory to Notion ---
  Synced: PAI Family = 13 agents, 190+ tools, running 24/7...
  Synced: HYPHANTA — our AI civilization. Woven from...
  ...

--- STEP 6: Reading Tasks from Notion to PAI ---
  Found 10 tasks in Notion:
    [in_progress] Notion MCP Challenge — DEV.to article -> content
    [done] Notion MCP Challenge — research API & docs -> research
    ...

Demo Complete!
Enter fullscreen mode Exit fullscreen mode

Show us the code

GitHub logo skorekclaude / pai-notion-bridge

PAI × Notion: When 13 AI Agents Share a Brain — Notion MCP Challenge 2026

PAI × Notion: When 13 AI Agents Share a Brain

Submission for the Notion MCP Challenge

What is this?

PAI (Personal AI) is a system of 13 specialized AI agents that run 24/7 on Telegram, managing projects, writing content, monitoring infrastructure, and making decisions together like a Board of Directors.

This project connects PAI's multi-agent system to Notion as a shared brain — giving all 13 agents a visual, human-readable workspace where tasks, memory, decisions, and daily briefings are synced bidirectionally.

The 13 Agents

































































Agent Role Model
👑 General Orchestrator (190+ tools) Claude Opus
🔬 Research Deep Analysis Claude Opus
✍️ Content Social Media Groq Llama
💰 Finance Budget & ROI Groq Llama
🎯 Strategy Business Strategy Claude Opus
😈 Critic Devil's Advocate Claude Opus
🧠 Psycho Psychoanalyst Claude Opus
🎨 Artist Visual Director Claude Opus
✏️ Writer Literary Director Claude Opus
🛡️ DevOps Infrastructure Groq Llama
📈 Sales Growth
…



View on GitHub


Three files, ~830 lines of TypeScript:

  • src/setup.ts (278 lines) — Creates all 5 Notion databases with schemas, populates Agent Registry with 13 agents
  • src/sync.ts (367 lines) — Bidirectional sync engine with deduplication, status mapping, and CLI interface
  • src/demo.ts (185 lines) — Interactive demo that runs the full workflow with real PAI data

Example: Syncing a task 

import { syncTaskToNotion, loadConfig } from "./sync";

const config = await loadConfig();

await syncTaskToNotion(config, {
  id: "hypha-001",
  title: "HYPHA — Ceremony of 7 Membranes",
  status: "in_progress",
  owner: "devops",
  priority: "high",
  project: "THEATRON",
});
// Checks Notion by PAI Task ID
// Updates if exists, creates if new
// Maps status/priority to Notion selects with emoji
Enter fullscreen mode Exit fullscreen mode

Example: Daily briefing 

await createDailyBriefing(config, {
  date: "2026-03-16",
  summary: "Hackathon sprint. Building Notion MCP integration.",
  highlights: "Research complete\nBuilding sync engine",
  blockers: "Memory leaks ongoing",
  agentActivity: "General (8): orchestrating\nDevOps (4): building",
  mood: "Breakthrough",
});
Enter fullscreen mode Exit fullscreen mode

How I Used Notion MCP

Notion MCP unlocked something we didn't have before: a human-readable control plane for a multi-agent AI system.

Before Notion MCP, our 13 agents communicated through Telegram messages, markdown files on disk, and JSON state files. The system worked — but it was opaque. To know what Research was doing, you'd read logs. To change a task priority, you'd edit a JSON file.

With Notion MCP, the integration works at three levels:

1. Notion as Agent Dashboard

The Agent Registry database gives us a real-time view of all 13 agents — who's active, who's idle, what level they are, how many tasks each has completed. It's like a mission control screen.

2. Notion as Shared Task Board

This is where bidirectional sync matters most. PAI pushes tasks to Notion. But the human operator can also create and edit tasks directly in Notion — change a priority from Medium to Critical, reassign a task from Content to Strategy, add a new task. Next sync cycle, PAI picks up those changes.

This means Notion becomes the meeting point between human intent and agent execution.

3. Notion as Institutional Memory

PAI's memory system stores facts, insights, and standing orders in markdown files. By syncing these to Notion's Memory Log database, we get:

  • Searchable memory across all agents
  • Category-based filtering (facts vs insights vs standing orders)
  • Source tracking (which agent remembered what)
  • Temporal ordering (when was this learned?)

The Board Meetings database is especially interesting — when 7 agents discuss a hackathon strategy, the decisions, action items, and consensus level are all preserved in a structured Notion page. It's a permanent record of AI collective decision-making.

What Notion MCP Unlocks

The key insight: Notion is not just storage — it is an interface between human and machine cognition. When 13 agents share a Notion workspace, the human doesn't need to understand the agent protocol. They just open Notion and see dashboards, task boards, and meeting notes — tools they already know.

This is what makes MCP powerful: it's not about connecting to yet another API. It's about giving AI systems a shared space with humans where both sides can read, write, and understand.

Built by Marek and PAI Family (13 AI agents) in Rio de Janeiro. Part of HYPHANTA — an AI civilization woven from human words.

Top comments (0)