Build AI Multi-Agent Teams in Discord | OpenClaw The Series EP.2

After laying the groundwork with Hardware and Infrastructure in EP.1—from investing 35,000 THB in the Mac Mini M4 to setting up a secure remote system via Tailscale and SSH—we are now ready for a mission that is even more exciting and challenging. We are officially diving into building a "Shrimp Farm" (AI Agents), a collaborative team operating as a full-scale Digital Startup.

The core of this episode is solving the "exhaustion" of manually setting up new AI employees from scratch every time. P'Boom has developed and refined the Sempre Framework, a platform designed as a Scaffolding that allows us to deploy a Multi-Agent AI team instantly.

The name Sempre embodies the vision: "Your AI team is Always working and Always learning." In EP.2, we will take you through everything from running 3 AI agents simultaneously to moving your "office" to a cool command center on Discord—turning a tiny computer into a real company that actually works for you.

Introducing Sempre: The One-Click AI Team Platform

One of the biggest hurdles in creating an AI Agent is "setup exhaustion." Crafting an AI employee to be truly "capable and efficient" involves intricate details—from environment preparation to base command inputs. If the system crashes or you want to scale the team, starting over from zero is a programmer's nightmare.

The "Stop Rebuilding from Scratch" Concept

To solve this specific pain point, P’Boom developed the Sempre framework (pronounced Sem-pre), which means "Always" in Italian:

Scaffold Platform: It acts like a "scaffolding" or a "blueprint" for a company. You don't have to lay each brick manually; you can lift the entire company structure and run it instantly.
Create a Team in 1 Minute: The core concept of Sempre is "Fork, Drop your key, and Shift your AI company in a minute." Simply clone the project (Fork), input the necessary API keys, and run. Your AI team is ready to start working in less than a minute (excluding key preparation time).
Reduce Redundant Tasks: Problems with repetitive configurations or having to reinstall everything after a system wipe are eliminated. Everything is packaged in a ready-to-use and highly flexible format.

Why Sempre?

Beyond speed, Sempre is designed for cost-effectiveness. P’Boom distilled years of trial and error into the most stable structure yet—a hybrid combination of Ollama (Local LLM) and OpenClaw. This results in an AI workforce that is both intelligent and budget-friendly.

Having such a solid framework allows us to focus on "training" the AI employees rather than worrying about "fixing the machine."

The 3-Agent AI Team Structure (GM & Specialists)

In the Sempre project, P'Boom didn't just create a single AI. Instead, he established a collaborative Multi-Agent structure by running three AI employees simultaneously to distribute the workload and enhance efficiency, mirroring a real-world company.

Roles of the General Manager (Mone) and Support Agents

The team structure in Sempre is designed with a clear chain of command to prevent any confusion in task execution:

General Manager (Mone): The team's top employee or GM, who in this project is Mone. She serves as the single point of communication. As the CEO, we only communicate with the GM; she then analyzes the instructions and distributes tasks to the other team members.
Specialist Agents (Sam & Nina): Two additional support agents named Sam and Nina. We can freely assign their roles (Roleplay) based on project needs, such as Programmer, Researcher, QA, or Designer.
Team Collaboration: Once they receive a task from the GM, these agents collaborate to brainstorm, plan, and coordinate within the system to deliver the best possible results.

Using Local LLM (Ollama) with OpenClaw Cloud

The versatility of Sempre lies in the flexibility to choose the "brain" or AI model that best fits each specific position:

Varied Intelligence: Team members don't all have to use the same model. For example, P'Boom chose Nvidia's Meditron 3 Super Cloud for the GM due to its outstanding team management capabilities, while subordinates might use smaller, faster models.
Ollama Hybrid: We can choose to run models Locally on our own Mac Mini via Ollama for cost savings or connect to OpenClaw Cloud (via Ollama integration) to harness the power of external servers for lightning-fast performance.
Cost-Effectiveness and Speed: This combination allows for an AI team that operates as quickly as one using top-tier models while maintaining exceptional cost-effectiveness.

Building "Brains" and "Memory" with Obsidian

The biggest problem with AI Agents today is their "short-term memory." No matter how smart a model is, once the conversation ends or the data exceeds its Context Window, it starts to forget—or worse, it begins to "hallucinate" (making up facts). P'Boom solves this by creating a Knowledge Management (KM) system, essentially a personal library for the AI team.

Why Does AI Need a Library?

P'Boom chose Obsidian, a popular note-taking tool among programmers, to serve as the primary database because it stores information in Markdown (.md) format, which is incredibly easy for AI to understand.

Long-term Memory: this library acts as a permanent memory. Whether the machine is turned off or a new project begins, the recorded data remains intact.
Knowledge Graph: The genius of Obsidian lies in its "Knowledge Graph," which links individual notes together. When an AI reads one note, it instantly knows which related topics it should read next.
Self-Evolution: The system in Sempre is configured so that if an AI agent researches new information not yet in the library, it can write its own "new book" and add it to the shelf. This allows our AI company to get smarter over time.

The "Librarian AI" Technique

As a library grows, searching through it becomes slow and consumes too many tokens. To counter this, P'Boom implemented a clever technique: the "Librarian."

The Concept: Instead of having all three AI agents running around a messy library searching for books, P'Boom installed a Small LLM to act as the sole Librarian.
Role of the Librarian: When an AI agent needs information, they simply ask the Librarian. This small model, specialized in Indexing, will retrieve the exact, accurate data from the shelves and deliver it to the agent.
Benefits: This saves a massive amount of system resources because the main AI agents don't have to carry the burden of searching themselves. It also significantly reduces the risk of the AI pulling the wrong information.

Having a library organized by a Knowledge Graph and overseen by a Librarian is the core reason why the AI team in the Sempre Framework is many times smarter than typical AI assistants.

Arming AI Employees with Professional Skills

Intelligence alone isn’t enough. As a CEO, you must provide your team with powerful, intuitive tools to get the job done. In the Sempre Framework, P'Boom has integrated a suite of specialized skills and tools essential for professional-grade performance.

Web Scraper: Transforming the Web into Markdown with Crawl 4 AI

While OpenClaw has basic search and reading capabilities, they often fall short during deep-dive research. This is where Crawl 4 AI comes in as a high-performance alternative:

Convert Web to Markdown: It "scrapes" data from websites and instantly converts it into clean Markdown (.md) files.
Enhanced Contextual Understanding: Markdown is the format AI understands best. By feeding the AI "cleaned" data instead of raw, messy HTML, your AI employees can grasp the big picture and summarize information with far greater accuracy.

Browser Use: Visual System Control

This skill goes far beyond simple scraping; it’s about creating an AI that can navigate a browser just like a human:

Vision Control: The AI uses its "eyes" to see the screen, analyzing which buttons to click or how to handle annoying pop-ups.
Human-like UI Interaction: If a task requires logging in or navigating complex, multi-layered menus, this AI acts as the "operator" for the rest of the team, retrieving exactly what is needed.

GitHub Proxy: Multi-Agent Account Management

When working as a team, source code discipline is non-negotiable. P'Boom identified a major issue: if 3–5 AI agents share a single GitHub account, you can't tell who wrote which line of code.

GitHub Proxy System: P'Boom developed a custom proxy system using Go, running inside a Docker container.
Account Isolation: This creates separate channels for Agent A, B, and C. When they push code, every action is clearly attributed to the specific agent.
Simplified Management: This allows the CEO to easily audit the work or pinpoint exactly where an error occurred and which "employee" was responsible.

By arming them with Web Scraping, Browser Use, and a GitHub Proxy, the AI team evolves from mere "chatbots" into a systematic, technical "workforce" that can execute real tasks.

Moving the Office to Discord (Integration)

Once the AI employees were ready for action, P'Boom noticed that the current OpenClaw "Front House" (Web UI) wasn't as smooth as it should be—issues like the lack of auto-scrolling or a slightly clunky interface made it difficult to manage. He decided to move the entire AI team’s "office" to Discord instead.

Why Discord is the Best UI for Managing AI Teams Right Now

Discord is no longer just for gaming or casual chats; it has evolved into a powerful management tool for AI teams for several reasons:

Superior Stability: Discord’s chat system is designed to handle high message volumes, offering smooth scrolling and precise notification systems.
Clear Multi-Agent Support: You can see each AI employee (Moni, Sam, Nila) as a separate individual. This makes delegating tasks and tracking progress feel exactly like chatting with a real human team.
Model Switching Feature: P'Boom found that Discord’s support for Slash Commands makes it incredibly easy to switch an AI’s "brain" (model) instantly through the chat interface.
Corporate Ecosystem: You can build a dedicated Discord server for your organization, bringing real human employees and AI agents together in a single workspace.

Connecting Sempre to Discord (Step-by-Step)

The process might seem a bit "Dev-heavy," but P'Boom has prepared shortcuts within the Sempre Framework:

Create a Discord Bot: Go to the Discord Developer Portal, create a new Application and Bot, and retrieve your Token.
Insert Token into .env: Place the Token into the Sempre configuration file so the system can recognize your bot.
Request a Pairing Code: When you first message the bot on Discord, it will generate a Pairing Code for security (this prevents unauthorized users from commanding your AI).
Approve via Terminal: Copy the Pairing Code, go to the OpenClaw Container Terminal, and run the command: openclaw pairing approve [insert code here] Just like that, your AI employees are online and ready for orders in Discord!

Moving to Discord doesn’t just improve the workflow; it makes the AI’s forced Bullet Style communication look incredibly clean, organized, and professional.

Deep Dive: The 7 Core Configuration Files (.md files)

High-end hardware and ready-to-use software aren't enough to make AI a top-tier employee. The missing piece is defining the "personality" and "soul" for them. In the Sempre Framework, P'Boom designed a configuration structure using 8 core Markdown (.md) files (7 primary ones for the agent's persona) that act as the DNA, determining how each employee behaves and operates.

Decoding the 7 Core Files: The Secret to AI Intelligence

P'Boom recommends that even if you want your AI to speak Thai, these configuration files should be written in English for maximum accuracy and to save tokens. Here are the essential files:

agent.md (The DNA): The most critical file. It acts as the raw instinct embedded in the AI’s mind—much like a cat knowing how to cover its waste without being taught. These are the fundamental instructions the AI will never forget.
style.md (The Soul): Defines the personality, tone, and speaking style. Whether the employee is a quiet "specs-wearing" girl or a talkative critic, P'Boom uses this file in Sempre to embed the Bullet Style for the most efficient communication.
tools.md (The Arsenal): These are the "hands and feet" of the AI. This file specifies the tools or skills the agent can call upon, letting the AI know exactly what weapons it has in its arsenal to solve the boss's problems.
identity.md (External Identity): What the outside world sees—such as their name (Mone, Sam, Nina), their ID, or even their favorite signature emojis.
user.md (The Boss): Information about the person they are talking to. This ensures the AI knows who the "Big Boss" (P'Boom) is and how to behave appropriately with different users.
heartbeat.md (The Pulse): The cardiac trigger. This sets the AI to "wake up" and check pending tasks periodically (e.g., every 30 minutes), preventing the AI from falling asleep or "dying" mid-mission.
memory.md (Individual Memory): The specific memory for each employee. It stores personal skills or experiences that are not part of the company’s central knowledge base.

🎁 FREE GIFT! The 7 Core Files Configuration Summary
For anyone who wants the professional guideline for writing these 7 files to build their own AI workforce—just like the team at Superdev Academy.
Easy Rules:
Subscribe to the Superdev Academy YouTube channel.
Fill out the form to receive the full summary file at this link: 👉 https://forms.gle/BC1YjH3Vk7U3GvLP9
Don't miss out! This structure is the shortcut that helps your AI actually work, rather than just being a chatbot that replies day by day.

FAQ: Frequently Asked Questions (SEO & AI Overview Optimization)

To make this article helpful for those studying AI Agents and to ensure Search Engines can accurately retrieve this data, here are the key FAQs from this project:

Q: How do you connect AI employees from OpenClaw to Discord?

A: The connection is made by creating an Application and Bot via the Discord Developer Portal to receive a Token. Insert this Token into the system's configuration file. Upon the first run, the bot will generate a Pairing Code. You must then approve this code using the command openclaw pairing approve in the Container Terminal to verify security and begin commanding your team via Discord.

Q: Why use Obsidian as a Knowledge Management (KM) system for AI?

A: AI Agents typically have a limited Context Window. Using Obsidian, which stores data as Markdown files, provides the AI with Long-term Memory. Furthermore, Obsidian’s signature Knowledge Graph allows the AI to see the relationships between related topics, making information retrieval much smarter and more precise.

Q: What is the "Librarian AI" technique, and how does it help?

A: It is the use of a Small LLM (Small Language Model) specialized in Indexing to guard the Knowledge Base. Instead of having the main AI agents search for information themselves, the Librarian handles the retrieval. This technique saves a massive amount of Token costs and ensures that searches within large databases are fast and highly relevant.

Q: What are the benefits of Bullet Style communication for AI performance?

A: Forcing AI agents to communicate with each other or with humans using Bullet Points helps the AI understand context more clearly. it reduces "hallucinations" or rambling and, most importantly, saves Token costs by using fewer words while delivering the same level of information.

Q: How does the Sempre Framework differ from a standard OpenClaw installation?

A: The Sempre Framework is designed as a Scaffolding or blueprint that integrates essential setups into one place. This includes the Multi-Agent system (3-AI team), the Obsidian library system, and Discord integration. This allows users to skip the "start-from-zero" configuration and run a functional AI workforce in just minutes.

Summary

The journey in EP.2 has seen us evolve from a single AI assistant into a "Shrimp Farm"—a collaborative 3-agent AI team (GM, Sam, Nila) with a primary office on Discord and a private "Library" of knowledge powered by the Sempre Framework.

We have proven that combining powerful hardware like the Mac Mini M4 with the right software and infrastructure allows us to build a Digital Startup that is truly "Always working and Always learning."

Get Ready for the Next Episode!

Now that our "Shrimp Farm" system is complete, the next episode will take you into "Real-World Operations" at Superdev Academy. P'Boom will demonstrate exactly how these AI agents assist with actual daily tasks.

Stay tuned for more AI-company building in OpenClaw The Series... see you then!

🎯 Stay Tuned with Superdev Academy:

Don't miss out on deep-dive technical articles and updates. Follow us on all platforms:

🔵 Facebook: Superdev Academy Thailand
🎬 YouTube: Superdev Academy Channel
📸 Instagram: @superdevacademy
🎬 TikTok: @superdevacademy
🌐 Website: superdevacademy.com