Skip to content

Ememzyvisuals/microdragon

Repository files navigation

Microdragon
  ███╗   ███╗██╗ ██████╗██████╗  ██████╗ ██████╗ ██████╗  █████╗  ██████╗  ██████╗ ███╗   ██╗
  ████╗ ████║██║██╔════╝██╔══██╗██╔═══██╗██╔══██╗██╔══██╗██╔══██╗██╔════╝ ██╔═══██╗████╗  ██║
  ██╔████╔██║██║██║     ██████╔╝██║   ██║██║  ██║██████╔╝███████║██║  ███╗██║   ██║██╔██╗ ██║
  ██║╚██╔╝██║██║██║     ██╔══██╗██║   ██║██║  ██║██╔══██╗██╔══██║██║   ██║██║   ██║██║╚██╗██║
  ██║ ╚═╝ ██║██║╚██████╗██║  ██║╚██████╔╝██████╔╝██║  ██║██║  ██║╚██████╔╝╚██████╔╝██║ ╚████║
  ╚═╝     ╚═╝╚═╝ ╚═════╝╚═╝  ╚═╝ ╚═════╝ ╚═════╝ ╚═╝  ╚═╝╚═╝  ╚═╝ ╚═════╝  ╚═════╝ ╚═╝  ╚═══╝

Distributed Intelligence Network Operating as Cognitive Local Autonomous Worker

npm License: MIT Status Core Lines Platform

Created by EMEMZYVISUALS DIGITALS Emmanuel Ariyo — Founder & CEO

npm · GitHub · @ememzyvisuals on X


The AI agent that actually does things.

You open your terminal. You type microdragon. You see:

  ⬡ MICRODRAGON  v0.1.0

  Hi! What do you want to achieve today?

No commands to memorise. No 47 flags. No docs to read first. Just describe your goal and MICRODRAGON handles the rest.

Three minutes after install, MICRODRAGON can:

  • 🔒 Audit your codebase for security vulnerabilities (OWASP Top 10, 8 attack patterns)
  • 🎮 Open GTA V and drive across Los Santos without you touching the keyboard
  • 🧠 Fine-tune an AI model on your own data (OpenAI API, LoRA local, or Ollama custom)
  • 📊 Read a 200-page PDF and answer precise questions about specific clauses
  • 💻 Generate, run, and self-debug code — only showing you the working result
  • 📈 Pull live stock data, compute RSI/MACD, and give a BUY/SELL/HOLD signal
  • 🤖 Act as your personal assistant — CEO briefing, developer standup, trader pre-market
  • 🔗 Remember everything across WhatsApp, Telegram, Discord, and CLI simultaneously
  • ⚡ Control Photoshop, Excel, Word — knows every panel, shortcut, and workflow
  • 🚀 Be your expert in Python, Rust, Go, React, AWS, K8s, and 40+ other technologies

No subscription. No cloud. Runs on your machine. Free with Ollama.

MICRODRAGON = Distributed Intelligence Network Operating as Cognitive Local Autonomous Worker


⚡ PUBLIC BETA v0.1.0 — Actively developed. Feature requests via @ememzyvisuals


Why MICRODRAGON Is Different

MICRODRAGON OpenClaw Claude Code Devin
Free option ✅ Ollama, forever ❌ $20+/month ❌ $20+/month
Fully local
Security (CVEs in 60 days) 0 9+
All domains Messaging/routing Code only Code only
Plays games ✅ CV + AI
Fine-tunes models ✅ 3 methods
Cybersecurity expert ✅ OWASP + SAST Partial
Personal assistant ✅ CEO/Dev/Trader
Cross-platform memory ✅ WhatsApp+TG+Discord Partial
HuggingFace models ✅ 12 verified
Self-debugs code ✅ 5-iteration loop Partial Partial
Source code ✅ MIT ✅ MIT
Skills/plugins ✅ Secure-scanned

Table of Contents

  1. System Requirements & Hardware Guide
  2. Install — Step by Step
  3. First Run
  4. AI Provider Setup
  5. HuggingFace Models
  6. Simple Mode vs Pro Mode
  7. Every Feature — Complete Reference
  8. Social Platform Setup
  9. Configuration Reference
  10. Environment Variables
  11. GitHub Actions — Deploy & Publish to npm
  12. Architecture
  13. Evaluation Results
  14. Contributing & Extensibility
  15. Limitations
  16. Troubleshooting

1. System Requirements & Hardware Guide

MICRODRAGON runs on ANY functional computer. No extra hardware required.

Hardware What you can run Model recommendation
512MB RAM, any CPU Basic MICRODRAGON (cloud AI) Groq free tier (no local model needed)
2GB RAM Full MICRODRAGON + local AI TinyLlama 1.1B (0.6GB)
4GB RAM Full MICRODRAGON + good local AI Phi-3 Mini 3.8B (2.3GB) — excellent quality
6-8GB RAM Full MICRODRAGON + great local AI Llama 3.1 8B (4.7GB) — RECOMMENDED
16GB RAM Full MICRODRAGON + near-GPT4 DeepSeek R1 14B (8.3GB)
32GB+ RAM or 8GB GPU Full MICRODRAGON + best open model Qwen 2.5 72B (41GB)

Key point: You do NOT need a GPU, a Mac Mini, a gaming PC, or any special hardware. A $30/month VPS, a 10-year-old laptop, or a Raspberry Pi 4 all run MICRODRAGON fully.

The only limitation is which local AI model fits in your RAM. With Groq (free tier, no download), even a $5 VPS gets full AI quality.

Exact requirements

Component Minimum Recommended
OS Windows 10 x64, macOS 12, Ubuntu 20.04 Windows 11, macOS 14, Ubuntu 22.04
CPU Any x64 or arm64 Any — more cores = faster self-debug
RAM 512MB (cloud AI) / 2GB (local AI) 8GB for best local model balance
Disk 500MB for MICRODRAGON 10GB if downloading local models
Node.js 18.0+ (required for install) 20 LTS
Python 3.9+ 3.11+
Internet For cloud AI / first install Not needed after Ollama model download

2. Install — Step by Step

Method A: npm (Recommended — 3 minutes)

# Step 1: Verify Node.js >= 18
node --version

# If not installed:
# Windows:  winget install OpenJS.NodeJS.LTS
# macOS:    brew install node
# Ubuntu:   curl -fsSL https://deb.nodesource.com/setup_20.x | sudo -E bash - && sudo apt install nodejs

# Step 2: Install MICRODRAGON
npm install -g @ememzyvisuals/microdragon

# Step 3: Verify
microdragon --version

# Step 4: Setup (choose AI provider, enter key)
microdragon setup

# Step 5: Start
microdragon

Method B: Build from Source

# Requires Rust 1.75+ and Python 3.9+
git clone https://github.com/ememzyvisuals/microdragon
cd microdragon

# Build Rust core (~3 minutes)
cd core && cargo build --release
cp target/release/microdragon ~/.local/bin/

# Install Python modules
cd .. && pip install -r requirements.txt
playwright install chromium

# Setup
microdragon setup

Windows Notes

MICRODRAGON works best in Windows Terminal (free from Microsoft Store):

winget install Microsoft.WindowsTerminal

Plain CMD works too — MICRODRAGON auto-detects CMD and uses plain text output.

Server / VPS Install

# Ubuntu 22.04 fresh server
apt update && apt install -y nodejs npm python3 python3-pip
npm install -g @ememzyvisuals/microdragon
microdragon setup
# Social bots (Telegram, Discord) run headlessly on any server

3. First Run

The first time you run microdragon, it shows a consent screen:

  ⬡ MICRODRAGON  v0.1.0-beta
  DISTRIBUTED INTELLIGENCE NETWORK OPERATING AS COGNITIVE LOCAL AUTONOMOUS WORKER
  by EMEMZYVISUALS DIGITALS — Emmanuel Ariyo

  ⚡ PUBLIC BETA — Please read before continuing

  ── DATA & PRIVACY ──────────────────────────────────────────────────────────
  • All data stored on YOUR machine at ~/.local/share/microdragon/
  • Encrypted with AES-256-GCM — we cannot decrypt it
  • API calls go to your configured provider (Anthropic/OpenAI/Groq/Ollama)
  • Zero telemetry, zero tracking, zero cloud sync

  ── PERMISSIONS ─────────────────────────────────────────────────────────────
  • MICRODRAGON only acts when you ask
  • Dangerous operations (delete, send email) require your confirmation
  • All actions are logged to ~/.local/share/microdragon/audit.log

  Type 'accept' to agree and continue, 'decline' to exit:

Type accept. This screen only appears once.


4. AI Provider Setup

MICRODRAGON needs an AI provider. Choose based on your needs:

Groq — Free tier, fastest (recommended for beginners)

# 1. Sign up at https://console.groq.com (no credit card)
# 2. Create API key → copy it (starts with gsk_)

microdragon config provider groq
microdragon config set-key groq gsk_YOUR_KEY_HERE

# Optional: choose model
microdragon config model llama-3.3-70b-versatile   # best quality
microdragon config model llama-3.1-8b-instant      # fastest

Anthropic Claude — Best quality

# https://console.anthropic.com → API Keys
microdragon config provider anthropic
microdragon config set-key anthropic sk-ant-YOUR_KEY_HERE
microdragon config model claude-sonnet-4-6

OpenAI

# https://platform.openai.com → API keys
microdragon config provider openai
microdragon config set-key openai sk-YOUR_KEY_HERE
microdragon config model gpt-4o

Ollama — 100% free, 100% offline

# 1. Install Ollama: https://ollama.ai/download
# 2. Pull a model
ollama pull llama3.1     # recommended: 4.7GB

# 3. Start Ollama
ollama serve

# 4. Configure MICRODRAGON
microdragon config provider custom
# Endpoint: http://localhost:11434/v1
# Model: llama3.1

Cost comparison

Provider Free Cost/1M tokens Speed Best for
Ollama (local) ✅ Forever $0 Medium Privacy, zero cost
Groq ✅ Free tier $0.05-$0.59 ⚡ Fastest Speed, free tier
Anthropic Haiku $0.25 Fast Quality + cheap
Anthropic Sonnet $3.00 Good Best quality
OpenAI GPT-4o $2.50 Good Strong coder

5. HuggingFace Models

Download any model from HuggingFace and use it as MICRODRAGON's brain.

# See all recommended models
microdragon model list

# MICRODRAGON detects your hardware and recommends the best model
microdragon model recommend

# Download via Ollama (easiest — handles everything)
microdragon model download phi3-mini         # 2.3GB — any computer
microdragon model download llama31-8b        # 4.7GB — recommended
microdragon model download deepseek-r1-14b   # 8.3GB — best reasoning
microdragon model download qwen25-72b        # 41GB — near GPT-4

# After download, configure MICRODRAGON to use it
microdragon config provider custom
# Endpoint: http://localhost:11434/v1
# Model: llama3.1  (or whatever you downloaded)

Verified Model List

Key Model Size RAM Best For Install
tinyllama TinyLlama 1.1B 0.6GB 1GB Basic tasks, ultra-low hardware microdragon model download tinyllama
phi3-mini Phi-3 Mini 3.8B 2.3GB 3GB Best small model, code+reasoning microdragon model download phi3-mini
llama31-8b Llama 3.1 8B 4.7GB 6GB Best all-rounder (RECOMMENDED) microdragon model download llama31-8b
mistral7b Mistral 7B 4.1GB 6GB Code, instruction following microdragon model download mistral7b
qwen25-7b Qwen 2.5 7B 4.3GB 6GB Multilingual, math microdragon model download qwen25-7b
codellama7b CodeLlama 7B 3.8GB 5GB Pure code tasks microdragon model download codellama7b
deepseek-r1-14b DeepSeek R1 14B 8.3GB 10GB Best reasoning, shows thinking microdragon model download deepseek-r1-14b
qwen25-14b Qwen 2.5 14B 8.7GB 11GB High quality all tasks microdragon model download qwen25-14b
llama31-70b Llama 3.1 70B 39GB 42GB Near GPT-4 quality microdragon model download llama31-70b
qwen25-72b Qwen 2.5 72B 41GB 44GB Best open model available microdragon model download qwen25-72b

Direct HuggingFace URLs are embedded in the model registry — see modules/huggingface/src/engine.py.


6. Simple Mode vs Pro Mode

Simple Mode (default when you type microdragon)

  ⬡ MICRODRAGON  Hi! What do you want to achieve today?

Just describe your goal. MICRODRAGON:

  1. Classifies your intent (10 goal categories)
  2. Asks ONE clarifying question if needed
  3. Shows what it's going to do
  4. Does it
  5. Asks what's next
  you → create a TikTok script about MICRODRAGON playing GTA

  Got it. I'll research viral AI gaming content, write a hook-first
  60-second script with timestamps, and build a hashtag pack.

  MICRODRAGON: Which character/mission should MICRODRAGON play?

  you → Scorpion on Dust2... wait wrong game. GTA V, drive with wanted level

  [60-second script, hooks, timestamps, 20 hashtags, thumbnail brief]

Commands in Simple Mode:

Command Effect
/pro Switch to full CLI Pro Mode
/help Show goal examples
/cost Session spend + tokens
/status Engine health + provider
/clear Clear conversation
/persona developer Set PA persona (developer/ceo/trader/gamer)
/exit Quit

Pro Mode

microdragon --pro   # start in Pro Mode
microdragon -p      # shorthand

7. Every Feature — Complete Reference


7.1 AI Conversation

microdragon ask "explain the CAP theorem with a real-world example"
microdragon ask --stream "write a complete SaaS business plan for an AI tool"
microdragon ask --agent coder "write a Redis rate limiter in Go"
microdragon ask --agent researcher "what are the biggest AI releases this month?"
microdragon ask --agent analyst "analyse the NVDA stock trend"
microdragon ask --agent writer "write a cold email to a Series A VC"
microdragon ask --agent security "audit this JWT implementation for vulnerabilities"
microdragon ask --output json "list top 5 Python async frameworks with pros/cons"

7.2 Code Generation & Debugging

# Generate complete, production-ready code
microdragon code generate "user authentication with JWT + refresh tokens in FastAPI"
microdragon code generate "Redis rate limiter using sliding window algorithm" --language go
microdragon code generate "real-time collaborative cursor system" --language typescript
microdragon code generate "binary search tree with in-order traversal" --language rust --output bst.rs

# Debug — MICRODRAGON reads your code, finds issues, shows exactly what to fix
microdragon code debug payment_service.py
microdragon code debug api/auth.ts
microdragon code debug main.rs

# Code review
microdragon code review src/
microdragon code review auth.py --security       # security-focused
microdragon code review . --recursive            # entire project

# Write tests
microdragon code test src/ --framework pytest
microdragon code test api.ts --framework vitest

# Git
microdragon code git "write a commit message for these staged changes"
microdragon code git "explain what this PR does" --pr 47

Self-debug in action:

$ microdragon code generate "web scraper with rate limiting" --self-debug

  Generating code...
  Running iteration 1... Error: aiohttp not installed
  [Self-Debug 2/5] Auto-fixing import...
  Running iteration 2... Error: AttributeError: 'NoneType' has no attribute 'select'
  [Self-Debug 3/5] Fixing selector...
  Running iteration 3... ✓ Success

  ─────────────────────────────────────────────────
  [Final working code — 3 iterations hidden from you]
  ─────────────────────────────────────────────────

7.3 Cybersecurity Expert

MICRODRAGON is a full cybersecurity expert — defensive and offensive (authorised testing only).

# Static code security analysis — scans for OWASP Top 10 and more
microdragon security audit payment_service.py
microdragon security audit src/              # entire directory
microdragon security audit . --recursive

# STRIDE threat modelling
microdragon security threat-model "SaaS API with user data and payments"
microdragon security threat-model "microservices app with JWT auth"

# OWASP Top 10 guidance on any topic
microdragon security owasp "SQL injection"
microdragon security owasp "SSRF"
microdragon security owasp "cryptography"
microdragon security owasp A01:2021          # specific OWASP ID

# Incident response playbooks
microdragon security incident "data breach"
microdragon security incident "ransomware"
microdragon security incident "compromised account"
microdragon security incident "DDoS"

# CVE lookup and guidance
microdragon security cve CVE-2021-44228      # Log4Shell guidance
microdragon security cve CVE-2023-44487      # HTTP/2 Rapid Reset

# Security checklist for your architecture
microdragon security checklist webapp
microdragon security checklist api
microdragon security checklist cloud

# HTTP security headers check
microdragon security headers https://yourdomain.com

# Compliance guidance
microdragon security compliance GDPR
microdragon security compliance SOC2
microdragon security compliance PCI-DSS

What it detects in code:

Vulnerability CWE Severity Example trigger
SQL Injection CWE-89 CRITICAL f"SELECT * FROM users WHERE id={user_id}"
Command Injection CWE-78 CRITICAL os.system("ls " + user_input)
Hardcoded Credentials CWE-798 HIGH password = "supersecret123"
XSS CWE-79 HIGH innerHTML = req.query.search
Path Traversal CWE-22 HIGH open(user_provided_path)
Weak Cryptography CWE-327 HIGH hashlib.md5(password)
SSRF CWE-918 HIGH requests.get(user_provided_url)
Insecure Deserialization CWE-502 CRITICAL pickle.loads(user_data)

Threat model output:

$ microdragon security threat-model "REST API with user auth and payments"

  STRIDE THREAT MODEL: REST API with user auth and payments

  SPOOFING        Risk: HIGH
    JWT token theft via XSS → impersonation
    Mitigate: httpOnly cookies, short JWT expiry, refresh token rotation

  TAMPERING       Risk: HIGH
    Parameter tampering (change price, user ID)
    Mitigate: Server-side validation, signed requests for payments

  REPUDIATION     Risk: MEDIUM
    Users deny making purchases
    Mitigate: Immutable audit log, digital receipts

  INFO DISCLOSURE Risk: HIGH
    Verbose errors exposing stack traces, PII in logs
    Mitigate: Generic error messages, PII scrubbing in logs

  DENIAL OF SERVICE Risk: HIGH
    Brute force login, large payload attacks
    Mitigate: Rate limiting, max payload size, CAPTCHA

  ELEVATION OF PRIVILEGE Risk: CRITICAL
    IDOR — user accesses other user's data
    Mitigate: Resource ownership check on every request

7.4 Web Research

microdragon research "best practices for multi-tenant SaaS 2025"
microdragon research "compare Supabase vs PlanetScale vs Neon" --sources 12
microdragon research "latest AI agent frameworks April 2026" --output comparison.md

7.5 Application Control

MICRODRAGON knows every panel, shortcut, workflow, and hidden feature of major applications.

Adobe Applications:

# Adobe Photoshop
microdragon apps photoshop "remove background from product.jpg"
microdragon apps photoshop "batch resize all images in ./photos/ to 1080x1080"
microdragon apps photoshop "create a dark Instagram post with text 'MICRODRAGON'"
microdragon apps shortcut photoshop "free transform"     # → Ctrl+T
microdragon apps shortcut photoshop "content aware fill" # → with selection

# Adobe Illustrator
microdragon apps illustrator "create a minimal tech logo for MICRODRAGON"
microdragon apps illustrator "export all artboards as SVG and PNG"
microdragon apps shortcut illustrator "pathfinder unite" # → Alt+click Unite

# Adobe Premiere Pro
microdragon apps premiere "cut every clip at scene changes"
microdragon apps premiere "add auto-captions to the sequence"
microdragon apps premiere "export for YouTube 1080p H.264"

# Adobe After Effects
microdragon apps aftereffects "animate this logo with a reveal effect"
microdragon apps aftereffects "create a text animation that writes itself"

# Adobe Lightroom, Audition, Acrobat
microdragon apps lightroom "apply cinematic grade to all images"
microdragon apps audition "remove background noise from interview.wav"
microdragon apps acrobat "fill all form fields in this PDF"

Microsoft Office:

# Excel
microdragon apps excel "create a monthly P&L spreadsheet with charts"
microdragon apps excel "analyse this sales.csv and find top 5 performers"
microdragon apps excel "add a VLOOKUP to match customer names to IDs"
microdragon apps shortcut excel "autosum"              # → Alt+=
microdragon apps shortcut excel "toggle absolute ref"  # → F4

# Word
microdragon apps word "create a legal NDA template"
microdragon apps word "format this document with professional headings"
microdragon apps shortcut word "heading 1"             # → Ctrl+Alt+1

# PowerPoint
microdragon apps powerpoint "create a 10-slide pitch deck for MICRODRAGON"
microdragon apps powerpoint "add speaker notes to each slide"

# Access, Outlook, OneNote
microdragon apps outlook "compose follow-up for all emails marked important"
microdragon apps onenote "create a project notes structure"

Other Applications:

microdragon apps vscode "open this project and run the test suite"
microdragon apps figma "export all components as PNG"
microdragon apps davinci "colour grade this video with a cinematic LUT"
microdragon apps obs "start streaming to Twitch"
microdragon apps blender "render this scene with Cycles"


7.5b Flyer & Poster Creation

When you say "create a flyer" or "make a poster", Microdragon does NOT guess what you want.

It asks questions first. Then downloads everything. Then designs.

you → Create a birthday flyer for my daughter's 18th birthday

  🐉 I can build that flyer for you.

  Here's exactly what happens:
  1. You answer a few quick questions (2 minutes)
  2. I download a background image matching your theme
  3. I download fonts matching your chosen style
  4. I load your logo or any photos you provide
  5. I open Photoshop and place everything automatically
     (or use my built-in renderer if Photoshop isn't installed)
  6. You get: PNG (share on WhatsApp) + PDF (print) + SVG (edit)

  Let's start:

  Whose birthday and how old are they turning?
  > Sofia's 18th

  Date and time of the celebration?
  > Saturday May 15 2026, 7pm

  Where is the party?
  > The Grand Ballroom, 14 Victoria Island, Lagos

  What vibe? (elegant / party / luxury / modern / neon)
  > luxury

  RSVP contact?
  > 08012345678

  Any photo or logo to include? (paste URL or describe)
  > (press Enter to skip)

  ──────────────────────────────────────────────────────────
  STEP 1 — Gathering assets...
  ✓ Background image downloaded  (luxury celebration lights)
  ✓ Font: Cormorant Garamond (found on system)
  ✓ Font: Raleway (downloaded from Google Fonts)
  ✓ Texture: luxury grain (downloaded)
  ✓ 4 assets ready in staging folder

  STEP 2 — Designing flyer...
  ✓ Photoshop found. Opening with ExtendScript...
  ✓ Photoshop script running — building flyer automatically
  ✓ Exporting PNG + PDF

  STEP 3 — Done!
  PNG: ~/Desktop/microdragon_flyers/flyer_final.png
  PDF: ~/Desktop/microdragon_flyers/flyer_final.pdf
  SVG: ~/Desktop/microdragon_flyers/flyer_final.svg
  ──────────────────────────────────────────────────────────

  Want to change anything? (colors, text, layout, style)

What Microdragon downloads before Photoshop opens:

  • Background/hero image (searches Unsplash by keyword automatically)
  • Fonts (from system font library or Google Fonts CDN)
  • Your logo (from URL or file path you provide)
  • Textures and overlays (based on chosen style)

Photoshop not installed? Microdragon uses its built-in SVG + Pillow renderer. Same design, no Photoshop needed.

Flyer styles: modern · elegant · bold · playful · corporate · luxury · party · neon

Flyer sizes: a4 (print) · a5 · square (social) · story (Instagram/WhatsApp) · banner

Commands:

# Natural language
microdragon
> Make a business flyer for my restaurant grand opening

# Pro mode
microdragon design flyer --type birthday --title "Sofia's 18th" \
  --date "May 15 2026" --venue "Grand Ballroom Lagos" --style luxury

# With your own images
microdragon design flyer --type business --image ./photo.jpg --logo ./logo.png

# WhatsApp story format
microdragon design flyer --type event --size story

See full documentation: docs/FLYER_CREATION.md


7.6 Game Simulation

# REQUIRED: pip install mss pynput opencv-python-headless numpy
# Game must be open, in focus, windowed or borderless-windowed mode

microdragon game list                                    # all supported games

# Open World
microdragon game play "GTA V"
microdragon game play "GTA V" --duration 600
microdragon game play "GTA San Andreas"
microdragon game play "Red Dead Redemption 2"

# Racing
microdragon game play "Need for Speed Heat"
microdragon game play "Forza Horizon 5"
microdragon game play "Gran Turismo 7"

# Fighting
microdragon game play "Mortal Kombat 11"
microdragon game play "Mortal Kombat 11" --character scorpion
microdragon game play "Mortal Kombat 1"
microdragon game play "Street Fighter 6" --character ryu
microdragon game play "Tekken 8"

# Shooters
microdragon game play "Counter-Strike 2"
microdragon game play "Call of Duty Modern Warfare"
microdragon game play "Apex Legends"
microdragon game play "Valorant"

# Session control
microdragon game stop
microdragon game pause
microdragon game status    # live: fps, actions/sec, health, accuracy estimate

Game knowledge — what MICRODRAGON knows:

For each supported game, MICRODRAGON has complete knowledge of:

  • All controls (keyboard and gamepad)
  • All cheat codes (GTA V: TURTLE for health, LAWYERUP to clear wanted, etc.)
  • Mechanics (GTA wanted system, NFS heat levels, MK Fatal Blow activation)
  • Strategies (PID lane-keeping for NFS, MK11 combo state machine, CS2 counter-strafe)
  • Hidden features (GTA stock market manipulation, NFS heat multiplier grinding)
  • Maps and locations (CS2 callouts for Mirage/Dust2/Inferno, GTA V money spots)
  • Character specifics (MK11 Scorpion spear B,F,1 — Sub-Zero ice ball D,F,1)

MICRODRAGON's game knowledge base covers: GTA V, GTA San Andreas, NFS Heat, Mortal Kombat 11, Mortal Kombat 1, Street Fighter 6, Tekken 8, Counter-Strike 2, Call of Duty, and Apex Legends.


7.7 Model Training & Fine-Tuning

# Check what's available on your hardware
microdragon train capabilities

# Build a training dataset from your MICRODRAGON conversations
microdragon train dataset build --source conversations --output training.jsonl
microdragon train dataset build --source files --path ~/Documents/ --output docs.jsonl

# Method 1: OpenAI fine-tuning (~$0.42 for 300 examples)
microdragon train fine-tune \
  --base gpt-3.5-turbo \
  --data training.jsonl \
  --name my-microdragon-expert \
  --provider openai

# Method 2: Local LoRA (free, needs 8GB VRAM)
pip install torch transformers peft trl bitsandbytes
microdragon train fine-tune \
  --base meta-llama/Llama-2-7b-hf \
  --data training.jsonl \
  --provider local

# Method 3: Ollama custom model (free, no GPU needed)
microdragon train fine-tune \
  --base llama3.1 \
  --name my-custom-microdragon \
  --provider ollama
# Then: ollama run my-custom-microdragon

# Check fine-tune job status
microdragon train status --job-id ft-abc123

7.8 Personal Assistant

MICRODRAGON adapts to your professional context. Set your persona once — it changes everything.

# Set your persona
microdragon config persona developer
microdragon config persona ceo
microdragon config persona trader
microdragon config persona gamer
microdragon config persona business

# Morning briefing (persona-aware)
microdragon assistant briefing

# Book a meeting
microdragon assistant book-meeting \
  --title "Q2 Strategy Review" \
  --attendees "cto@company.com,cpo@company.com" \
  --duration 60 \
  --times "Monday 14:00,Tuesday 10:00,Wednesday 15:00"

# Prepare for a meeting
microdragon assistant prep-meeting "Board presentation" --attendees "board@company.com"

# Follow-up email after meeting
microdragon assistant follow-up \
  --meeting "Q2 Strategy" \
  --actions "Sarah to deliver roadmap by Friday,John to hire 2 engineers by month end"

# Daily plan
microdragon assistant plan-day

Developer PA morning briefing:

$ microdragon assistant briefing  [persona: developer]

  📅 MORNING BRIEFING — Thursday, April 11 2026

  TODAY'S FOCUS:
  1. Finish auth module (due Thursday)
  2. Review 3 open PRs in microdragon-skills repo
  3. Fix CI failure on main branch

  PRs AWAITING REVIEW:
  • #47 — feat: add Notion skill (waiting 2 days)
  • #51 — fix: webhook timeout (critical, blocking deployment)

  CI/CD STATUS:
  ⚠ Build failing on main — pytest import error since yesterday

  SCHEDULE:
  10:00  Team Standup (15 min)
  14:00  Architecture review with CTO (45 min)

  TECH NEWS:
  • OpenAI released GPT-5 preview to select developers
  • Rust 2.0 roadmap published

Trader PA morning briefing:

$ microdragon assistant briefing  [persona: trader]

  📈 PRE-MARKET — Thursday, April 11 2026  [05:45 EDT]

  OVERNIGHT:
  • BTC: $71,400 (+2.3%) — broke above resistance at $70k
  • S&P500 Futures: +0.4% — positive CPI data yesterday
  • NVDA: +3.1% AH after analyst upgrade

  WATCHLIST:
  AAPL  ▲ $184.32  RSI: 61  Signal: WATCH
  NVDA  ▲ $897.54  RSI: 74  Signal: OVERBOUGHT — take profits
  BTC   ▲ $71,400  RSI: 68  Signal: HOLD — momentum strong

  TODAY'S CALENDAR:
  08:30  Initial Jobless Claims
  10:00  Fed Speaker: Williams
  NVDA Earnings: Monday after close (watch IV)

  ⚠ Not financial advice.

7.9 Cross-Platform Memory

MICRODRAGON remembers across ALL platforms. Start a conversation on WhatsApp, continue on Telegram, reference it from CLI.

# Link your accounts (done once per platform)
microdragon memory link --platform telegram --id your_telegram_username

# View your conversation history across platforms
microdragon memory history --limit 20
microdragon memory history --platform whatsapp

# Your profile
microdragon memory profile

# Clear history on a specific platform
microdragon memory clear --platform telegram

# Clear all history
microdragon memory clear --all

How it works:

You send on WhatsApp: "remind me about the pitch tomorrow"
   → stored in unified_memory.db: {user, platform: "whatsapp", content, timestamp}

You then say on Telegram: "what did I say about the pitch?"
   → MICRODRAGON fetches last 20 messages across ALL platforms
   → Sees your WhatsApp message
   → "You mentioned on WhatsApp that you have a pitch tomorrow."

You type on CLI: "draft my intro for the pitch"
   → Full context from WhatsApp + Telegram is in the AI's context window
   → Generates pitch intro using your full history

7.10 Documents & Files

# Ask questions about any document
microdragon document read contract.pdf "what are the payment terms and penalties?"
microdragon document read annual_report.pdf "what was net profit 2025 vs 2024?"
microdragon document read terms.pdf "does this allow commercial use of my data?"
microdragon document read codebase.py "explain this code step by step"
microdragon document read employees.xlsx "who has been employed the longest?"

# Create documents
microdragon document create "Product Requirements Document for MICRODRAGON v1.0"
microdragon document create "Employment contract template UK law" --style legal

# Summarise a folder of documents
microdragon document summarise ~/Downloads/
microdragon document summarise ~/Projects/docs/

# Semantic file search (by meaning, not keywords)
microdragon files search "authentication and security" ~/Projects/
microdragon files search "revenue projections Q3" ~/Documents/

7.11 Browser Automation (experimental)

microdragon automate browser "go to news.ycombinator.com, get the top 10 stories"
microdragon automate browser --url "https://linkedin.com" "find all open Rust engineer roles"
microdragon automate browser "screenshot my portfolio site" --headless
microdragon automate browser "fill the contact form on example.com with test data"

# Scheduled automation
microdragon automate schedule "check competitor pricing every Monday 9am" --cron "0 9 * * 1"

7.12 Voice (experimental)

microdragon voice listen                      # record 5 seconds, transcribe, respond
microdragon voice listen --duration 15        # 15 seconds
microdragon voice say "Good morning. You have 3 meetings today."
microdragon voice transcribe meeting.mp3
microdragon voice transcribe lecture.wav --output transcript.txt
microdragon voice wake                        # always-on "hey microdragon" wake word

7.13 Email (experimental)

MICRODRAGON NEVER sends email without showing full preview + requiring you to type yes.

# Setup: Gmail App Password
export MICRODRAGON_EMAIL="you@gmail.com"
export MICRODRAGON_EMAIL_PASSWORD="your_16_char_app_password"

microdragon email inbox                       # summarise unread
microdragon email inbox --limit 50
microdragon email search "invoice"
microdragon email read <uid>

# Compose — always shows preview, always requires yes/no
microdragon email compose \
  --to ceo@bigcompany.com \
  --subject "Partnership Proposal" \
  --body "Hi, I'd like to propose..."

7.14 Calendar & Meetings (experimental)

microdragon calendar today
microdragon calendar week
microdragon calendar briefing

# Create events (confirmation required)
microdragon calendar add "Board Meeting" "2026-04-15 10:00" --duration 120 --attendees "board@company.com"
microdragon calendar add "1:1 with Sarah" "tomorrow 14:00" --location "Zoom"

7.15 GitHub Integration (experimental)

export GITHUB_TOKEN="ghp_your_token"

microdragon github review https://github.com/owner/repo/pull/47          # AI review
microdragon github review https://github.com/owner/repo/pull/47 --post   # post to GitHub
microdragon github issue ememzyvisuals/microdragon "Bug: game crashes on macOS"
microdragon github overview ememzyvisuals/microdragon

7.16 Business & Finance

microdragon business market AAPL
microdragon business market BTC-USD --interval 1h
microdragon business signal NVDA --detailed    # RSI, MACD, trend, BUY/SELL/HOLD
microdragon business signal SPY
microdragon business risk TSLA
microdragon business portfolio

7.17 Product Strategy

microdragon strategy prd \
  --product "MICRODRAGON Skills Marketplace" \
  --users "developers who want to extend MICRODRAGON" \
  --problem "users want custom capabilities but can't write Rust"

microdragon strategy market "AI agent software 2026"
microdragon strategy roadmap \
  --product "MICRODRAGON" \
  --current "v0.1.0 beta, 115 files, local-first" \
  --goals "1M npm installs, viral adoption"

7.18 Skills & Plugins

MICRODRAGON is extensible. Anyone can write a skill in Python and add it.

microdragon skills list
microdragon skills install https://github.com/someone/microdragon-weather
microdragon skills install ./my_skill/
microdragon skills run weather "Lagos"
microdragon skills scan ./my_skill/            # security scan before install
microdragon skills remove weather

Writing your own skill — minimal example:

Create this structure:

my_skill/
├── SKILL.md        (metadata)
└── main.py         (code)

SKILL.md:

# Skill: CryptoPrice
**Author:** your_username
**Version:** 1.0.0
**Type:** data
**Description:** Real-time crypto prices
**Commands:**
  - crypto price <symbol>
**Permissions:** network

main.py:

from microdragon.skills import SkillBase, command, SkillResult
import aiohttp

class CryptoPriceSkill(SkillBase):
    name = "crypto"

    @command("price")
    async def get_price(self, symbol: str, **kwargs) -> SkillResult:
        async with aiohttp.ClientSession() as s:
            async with s.get(
                f"https://api.coingecko.com/api/v3/simple/price"
                f"?ids={symbol}&vs_currencies=usd"
            ) as r:
                data = await r.json()
        price = data.get(symbol, {}).get("usd", "not found")
        return SkillResult(
            success=True,
            output=f"{symbol.upper()}: ${price:,.2f}" if isinstance(price, float) else price
        )

Install:

microdragon skills install ./my_skill/
microdragon skills run crypto price bitcoin
# → BITCOIN: $71,400.00

Skill types:

  • data — fetch/process external data (prices, weather, sports)
  • automation — control apps, fill forms, click buttons
  • ai — custom AI workflows using MICRODRAGON's brain
  • integration — connect to external APIs (Slack, Notion, Airtable)
  • research — web intelligence, OSINT, aggregation
  • utility — transform data (QR, unit convert, format)

Full guide: modules/skills/SKILLS_GUIDE.md


7.19 Social Platforms

microdragon social whatsapp start     # scan QR with phone
microdragon social whatsapp stop
microdragon social telegram start
microdragon social discord start
microdragon social status             # check all platforms

7.20 Background Watch

microdragon watch "alert me when AAPL drops below $150"
microdragon watch "notify on Telegram when email from investor@vc.com arrives"
microdragon watch "run daily market briefing at 8am weekdays"
microdragon watch "alert if disk usage exceeds 90%"
microdragon watch list
microdragon watch remove 3

7.21 Tech Skills

MICRODRAGON is an expert in every technology. Ask anything:

microdragon ask "write a K8s HPA config for a FastAPI service"
microdragon ask "what is the difference between process isolation and container isolation?"
microdragon ask "write a Terraform module for ECS Fargate with ALB"
microdragon ask "implement JWT refresh token rotation in Go"
microdragon ask "explain the SAGA pattern for distributed transactions"
microdragon ask "write a PostgreSQL query that finds the N most recent orders per customer"
microdragon ask "how do I detect SQL injection in a code review?"

# Full tech skills list
microdragon skills tech list

Domains with expert-level knowledge:

Languages: Python, Rust, JavaScript/TypeScript, Go, Java, Kotlin, Swift, Dart, C/C++, C#, PHP, Ruby, Scala, Elixir, Haskell, Shell, SQL, Solidity

Web: React, Next.js, Vue, Nuxt, Angular, Svelte, FastAPI, Django, Express, NestJS, Spring Boot, Laravel, Rails, Phoenix, Axum, Actix-Web

Mobile: React Native, Expo, Flutter, iOS (SwiftUI), Android (Jetpack Compose)

Cloud: AWS (200+ services), GCP, Azure, Vercel, Cloudflare Workers, Fly.io

DevOps: Docker, Kubernetes, Helm, ArgoCD, Terraform, Ansible, Pulumi, GitHub Actions, GitLab CI, Prometheus, Grafana, OpenTelemetry

Databases: PostgreSQL, MySQL, MongoDB, Redis, DynamoDB, Firestore, Cassandra, Elasticsearch, InfluxDB, Neo4j, pgvector, Pinecone, Weaviate

AI/ML: PyTorch, HuggingFace, LangChain, LlamaIndex, LoRA/QLoRA, vLLM, MLflow, W&B

Security: OWASP, SAST, Penetration Testing, Cryptography, OAuth2/OIDC, GDPR, SOC 2

Architecture: Microservices, Event-Driven, CQRS, DDD, Hexagonal, Distributed Systems


8. Social Platform Setup

WhatsApp

# Requirements: Node.js 18+
microdragon social whatsapp start
# Scan QR: WhatsApp → Settings → Linked Devices → Link Device

Commands in WhatsApp: Send any message → AI responds. /microdragon help for command list.

Telegram

# 1. @BotFather → /newbot → create → copy token
microdragon config set-key telegram 1234567890:TOKEN
microdragon social telegram start
# Open your bot in Telegram → /start

Discord

# 1. discord.com/developers → New App → Bot → Reset Token → copy
# 2. Enable: Message Content Intent + Server Members Intent
# 3. OAuth2 → URL Generator → bot → Send Messages → invite to server
microdragon config set-key discord YOUR_BOT_TOKEN
microdragon social discord start
# In server: !microdragon hello

9. Configuration Reference

Config at: ~/.config/microdragon/config.toml

[ai]
active_provider = "groq"
temperature = 0.7

[security]
sandbox_execution = true
audit_log_enabled = true
encryption_enabled = true

[assistant]
persona = "developer"          # developer/ceo/trader/gamer/business
user_name = "Emmanuel"
timezone = "Africa/Lagos"

[storage]
database_path = "~/.local/share/microdragon/microdragon.db"
unified_memory_path = "~/.local/share/microdragon/unified_memory.db"

10. Environment Variables

# AI Providers
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export GROQ_API_KEY="gsk_..."
export OPENROUTER_API_KEY="sk-or-..."

# Social
export TELEGRAM_BOT_TOKEN="..."
export DISCORD_BOT_TOKEN="..."
export GITHUB_TOKEN="ghp_..."

# Email
export MICRODRAGON_EMAIL="you@gmail.com"
export MICRODRAGON_EMAIL_PASSWORD="app_password"

# Voice
export ELEVENLABS_API_KEY="..."

# Cost control
export MICRODRAGON_PREFER_CHEAP="true"
export MICRODRAGON_PREFER_LOCAL="true"
export MICRODRAGON_DAILY_BUDGET_USD="1.00"

# PA settings
export MICRODRAGON_PERSONA="developer"
export MICRODRAGON_USER_NAME="Emmanuel"

# Display
export NO_COLOR="1"
export MICRODRAGON_LOG="info"

11. GitHub Actions — Deploy & Publish to npm

How it works

git tag v0.1.0 && git push --tags
     ↓
GitHub Actions runs automatically (~15 minutes)
     ↓
1. Tests on Ubuntu + macOS + Windows
2. Builds Rust binary for 5 platforms in parallel:
   Linux x64, Linux ARM64, macOS Intel, macOS ARM64, Windows x64
3. Creates GitHub Release with all binaries + SHA256SUMS
4. Publishes @ememzyvisuals/microdragon to npm
     ↓
npm install -g @ememzyvisuals/microdragon

One-Time Setup

Step 1 — Create GitHub repo:

git init && git add .
git commit -m "feat: MICRODRAGON v0.1.0 — Distributed Intelligence Network"
git remote add origin https://github.com/ememzyvisuals/microdragon.git
git push -u origin main

Step 2 — Get npm token:

# npmjs.com → Profile → Access Tokens → Generate New Token → Automation
# Copy: npm_xxxxxxxxxxxx

Step 3 — Add secret to GitHub:

github.com/ememzyvisuals/microdragon
  → Settings → Secrets → Actions → New repository secret
  Name: NPM_TOKEN
  Value: npm_xxxxxxxxxxxx

Step 4 — Enable Actions permissions:

Settings → Actions → General → Workflow permissions
  → ✅ Read and write permissions

Release

git tag v0.1.0
git push origin v0.1.0
# Done. GitHub Actions handles everything.

Version guide: bug fix → 0.1.1, new feature → 0.2.0, breaking → 1.0.0

Monitor:

open https://github.com/ememzyvisuals/microdragon/actions
npm view @ememzyvisuals/microdragon   # verify after ~15 minutes

12. Architecture

microdragon/ (115 files, 24,000+ lines)
│
├── .github/workflows/
│   ├── release.yml     5 platforms → GitHub Release → npm publish
│   └── ci.yml          Rust + Python + npm checks on every PR
│
├── core/src/           Rust engine
│   ├── main.rs         Entry: Simple Mode → API → watch daemon
│   ├── engine/         NexaEngine, 7 agents, 9-phase pipeline, TokenGuard
│   ├── brain/          AI routing, cost optimization, streaming
│   ├── cli/            Simple Mode UX + Pro Mode CLI
│   ├── memory/         SQLite (7 tables, WAL) + Vector (cosine similarity)
│   ├── security/       AES-256-GCM, injection guard, sandbox, audit
│   └── watch/          Background daemon + condition evaluation
│
├── modules/            Python capabilities
│   ├── gaming/         OpenCV vision + strategies + MK combo library + PID
│   │   └── knowledge/  Complete game knowledge: GTA V, NFS, MK11, CS2, COD
│   ├── apps/           Photoshop, Illustrator, Premiere, Excel, Word control
│   │   └── knowledge/  Complete UI/shortcut/workflow knowledge for all apps
│   ├── security_expert/ OWASP SAST scanner, threat modelling, IR playbooks
│   ├── training/       Fine-tuning (OpenAI+LoRA+Ollama), self-debug, strategy
│   ├── assistant/      PA engine, cross-platform memory, persona system
│   ├── huggingface/    Model downloader, 12 verified models with direct URLs
│   ├── tech_skills/    Expert knowledge: 40+ languages, cloud, DevOps, ML
│   ├── skills/         Plugin system with security scanning
│   ├── voice/          Whisper STT, TTS, wake word
│   ├── email/          IMAP/SMTP with confirmation gate
│   ├── social/         WhatsApp + Telegram + Discord
│   └── [research/coding/business/design/document/github/calendar]
│
├── npm/                @ememzyvisuals/microdragon package
├── eval/               60-test static benchmark + runtime results
└── docs/               Creation documentation + future roadmap

13. Evaluation Results

MICRODRAGON ships with a benchmark suite. Run it yourself:

python3 eval/benchmarks/microdragon_eval.py

Static (60 tests): 60/60 PASS — all architecture, security, pipeline, agent, memory, cost, game, training, social, CI/CD, UX, and long-horizon tests verified.

Runtime (65 tests): 52/65 (80%) — all logic tests including security guard (8/8), game strategies (6/6), goal classification (11/12), agent routing (8/10).

1 genuine bug found and fixed: SQLite DEFAULT clause portability issue.


14. Contributing & Extensibility

MICRODRAGON is MIT licensed and designed to be forked and extended.

git clone https://github.com/ememzyvisuals/microdragon
cd microdragon/core && cargo test
python3 eval/benchmarks/microdragon_eval.py   # run benchmarks

Extension Points

What you want Where to add it
New skill/capability modules/skills/ — write a Python skill
New application knowledge modules/apps/knowledge/app_knowledge_base.py
New game knowledge modules/gaming/knowledge/game_knowledge_base.py
New AI provider core/src/brain/model_router.rs
New agent type core/src/engine/agents.rs
New language in tech skills modules/tech_skills/src/engine.py
New social platform modules/social/
New security detection modules/security_expert/src/engine.py

For contributors

  1. Fork the repo
  2. Create a feature branch
  3. Run python3 eval/benchmarks/microdragon_eval.py — must stay 60/60
  4. Add a test for your feature in nexa_eval.py
  5. Open a PR

Questions: @ememzyvisuals on X


15. Limitations

Limitation Status
Self-debug only fixes shallow errors Known — structural repair is v0.2.0
Rust binary not yet in GitHub Releases Needs first tag push
Game accuracy is algorithm-target, not measured Live testing needs real game environment
No end-to-end integration tests v0.2.0 milestone
Social bots require manual token setup Expected — documented above

16. Troubleshooting

Problem Solution
microdragon: command not found export PATH="$PATH:$(npm root -g)/../.bin"
"Not configured" microdragon setup or set GROQ_API_KEY
Python modules missing pip install -r requirements.txt && playwright install chromium
Slow responses microdragon config provider groq — free + fastest
Game not responding Game must be in windowed mode and in focus
WhatsApp QR not showing cd modules/social/node_bridge && npm install
npm publish 403 npm login --scope=@ememzyvisuals
Windows CMD broken Install Windows Terminal from Microsoft Store
Build fails rustup update stable && cd core && cargo build --release

License

MIT © 2026 EMEMZYVISUALS DIGITALS


⬡ MICRODRAGON

The AI agent that actually does things.

Local-first. Rust core. Every skill. Every platform. Every domain.

EMEMZYVISUALS DIGITALS — Emmanuel Ariyo, Founder & CEO

npm · GitHub · @ememzyvisuals · Website

Star this repo if MICRODRAGON helped you. Every star makes the project more discoverable.
⭐ → github.com/ememzyvisuals/microdragon

About

🐉 CLI-first AI agent. Local. Secure. Multi-domain.

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors