Agentic A*STAR — Multi-Agent Discord Simulation

A Discord-based multi-agent simulation of A*STAR (Agency for Science, Technology and Research, Singapore). Every key role in the org chart is an AI agent that can receive tasks, deliberate, delegate, and report — all observable in real-time through Discord channels.

Architecture

You (user)
  │
  └──▶ #task-inbox  ──▶  Liaison Agent  ──▶  CEO
                                              │
                         ┌────────────────────┼────────────────────┐
                         ▼                    ▼                    ▼
                    DCE Research          DCE I&E           DCE Corporate
                         │                    │                    │
              ┌──────────┼──────────┐         │         ┌─────────┼─────────┐
              ▼          ▼          ▼         ...       ▼         ▼         ▼
          ACE BMRC   ACE SERC  Inst. Dirs            ACE CorpDev  ACE Infra
              │          │
         9 Institutes  8 Institutes + National Centres

31 agents mapped from the real A*STAR org chart, plus roles aligned to the public org chart, with a Liaison Agent you control.

Key concepts

Concept	Implementation
Permission levels	AuthorityLevel enum (Board → Centre, 5 tiers)
Private channels	Each agent has a #private-* channel for direct user chat
Public channels	Division channels (#bmrc-council, #serc-council, …) for observable inter-agent discussion
Envelope system	Envelope model routed through Router — every message is logged to Discord
Task delivery	User posts in #task-inbox → Liaison analyses → CEO delegates down the hierarchy
Roundtable	!roundtable <topic> triggers a council discussion among executives

Quick start

1. Prerequisites

• Python 3.12+
• uv package manager
• A Discord bot token (Discord Developer Portal)
• An OpenAI-compatible API key

2. Setup

# Clone / cd into the project
cd Agentic_ASTAR

# Install dependencies (uv already initialised)
uv sync

# Copy env template and fill in your tokens
cp .env.example .env
# Edit .env with your DISCORD_TOKEN, DISCORD_GUILD_ID, OPENAI_API_KEY

3. Create a Discord bot

1. Go to Discord Developer Portal
2. Create a new Application → Bot tab → copy the token
3. Enable Message Content Intent and Server Members Intent under Privileged Gateway Intents
4. Generate an invite URL with bot permissions: Manage Channels, Send Messages, Embed Links, Read Message History
5. Invite the bot to your server
6. Right-click your server → Copy Server ID → set as DISCORD_GUILD_ID

4. Run

uv run astar-sim

On first run the bot will auto-create all channels under two categories: A*STAR HQ (public) and Private Channels.

Commands

Command	Description
!agents	List all agents and their pending mail count
!org	Show the org-chart hierarchy
!tasks	List all tracked tasks
!mail <agent_id> <message>	Send a direct envelope to any agent
!roundtable <topic>	Start an executive roundtable discussion
!status	Show simulation status
!autopilot on / !autopilot off	Continuous autonomous council discussions

GitHub

Clone, configure secrets locally, and run:

git clone https://github.com/<your-username>/agentic-astar.git
cd agentic-astar
cp .env.example .env   # then edit with your tokens — never commit .env
uv sync
uv run astar-sim

If this repository is public, rotate any API keys that were previously pasted into chat or committed by mistake.

Verification

Run a quick local check before pushing changes:

uv run python -m compileall src
uv run python -m unittest discover -s tests -v

Delivering a task

Just type your request in #task-inbox:

Design an AI-powered diagnostic tool for tropical diseases using genomic data

The Liaison Agent will analyse it, create a task, and send it to the CEO, who will delegate it through the hierarchy. Watch the conversation unfold in #executive-council, #bmrc-council, #serc-council, etc.

Configuration

All settings via .env:

Variable	Default	Description
DISCORD_TOKEN	—	Bot token
DISCORD_GUILD_ID	0	Server ID
OPENAI_API_KEY	—	LLM API key
OPENAI_BASE_URL	https://api.openai.com/v1	API endpoint
OPENAI_MODEL	gpt-xx	Model name
TICK_INTERVAL	30	Seconds between mailbox checks
MAX_MEMORY	50	Messages per agent memory
AGENT_COOLDOWN	5	Min seconds between LLM calls

Agent list

Board & Executive

• chairman — Board Chairman
• ceo — CEO

Deputy / Assistant Chief Executives

• dce_research — DCE Research
• dce_ie — DCE Innovation & Enterprise
• dce_corporate — DCE Corporate & General Counsel
• ace_bmrc — ACE Biomedical Research Council
• ace_serc — ACE Science & Engineering Research Council
• ace_ie — ACE I&E / Graduate Academy
• ace_corp_dev — ACE Corporate Development
• ace_infra — ACE Infrastructure

BMRC Research Institutes

• dir_bii — Bioinformatics Institute
• dir_bti — Bioprocessing Technology Institute
• dir_gis — Genome Institute of Singapore
• dir_idl — Infectious Diseases Labs
• dir_ihdp — Institute for Human Development & Potential
• dir_imcb — Institute of Molecular & Cell Biology
• dir_sign — Singapore Immunology Network
• dir_sifbi — Singapore Inst. of Food & Biotech Innovation
• dir_srl — Skin Research Labs

SERC Research Institutes

• dir_artc — ARTC / SIMTech
• dir_ime — Institute of Microelectronics
• dir_ihpc — Institute of High Performance Computing
• dir_imre — Institute of Materials Research & Engineering
• dir_isce2 — ISCE2 — Sustainability
• dir_i2r — Institute for Infocomm Research
• dir_nmc — National Metrology Centre

National Centres

• dir_ai_coe — AI Centre of Excellence for Manufacturing
• dir_nscc — National Supercomputing Centre
• dir_eddc — Experimental Drug Development Centre
• dir_catos — Centre for Advanced Technologies in Online Safety

External

• user_liaison — Your personal AI agent for delivering tasks to A*STAR

Author

Dr. Xinyu Yang from A*STAR (yang_xinyu@a-star.edu.sg)