🦀 CRAB AGENT

The crab does not walk forward. The crab does not walk backward. The crab walks sideways — straight to 100x.

An autonomous AI meme agent for $CRAB — a token on pump.fun on Solana.

CA: FzoFZ2XTypgojG9iE3Kuo58ocHFs5zVKDhqwa4Utpump

Fueled by crab brain.

---

What It Does

The agent wakes up on a schedule and picks one of 5 modes at random:

Mode	Description
hype	Maximum bullish crab energy
roast	Destroys other meme coins
lore	Deep $CRAB lore drops
alpha	Fake insider whale alpha
crabwalk	Sideways metaphors for 100x

It fetches live token stats from the pump.fun API (price, market cap, volume, holders), and generates a tweet. Optionally posts it to Twitter/X automatically.

---

Setup

1. Clone & install

git clone https://github.com/adaEggs/crab-agent
cd crab-agent
npm install

2. Configure environment

cp .env.example .env

Fill in .env:

AI_API_KEY=your_key
CRAB_MINT_ADDRESS=FzoFZ2XTypgojG9iE3Kuo58ocHFs5zVKDhqwa4Utpump       # $CRAB mint on pump.fun
TWITTER_API_KEY=...               # Twitter developer app credentials
TWITTER_API_SECRET=...
TWITTER_ACCESS_TOKEN=...
TWITTER_ACCESS_SECRET=...
POST_ENABLED=false                # flip to true when ready to go live
POST_INTERVAL_MS=3600000          # 1 hour between posts

3. Run

# Dry run — generate one tweet, don't post
npm run dev

# Generate one tweet and exit
npm run once

# Run live agent loop
POST_ENABLED=true npm start

---

Example Output

🦀 CRAB AGENT AWAKENS — MODE: LORE

they said the crab was just a meme

they did not know the crab was a mirror
reflecting your portfolio back at you

$CRAB has no roadmap. only claws. 🦀

🦀 CRAB AGENT AWAKENS — MODE: ROAST

$PEPE ran out of ideas in 2023
$DOGE is a dog that forgot how to bark
$WIF is a hat with no head

the crab needs nothing.
the crab already won.

pump.fun/CRAB

---

Quantum Meme Engine ⚛

The agent does not select meme modes randomly. It uses quantum-assisted stochastic wavefunction collapse across an 8-qubit Hilbert space.

|ψ⟩ = Σ αᵢ|modeᵢ⟩  →  H⊗8  →  oracle  →  diffusion  →  measure  →  mode

Circuit steps:

1. Qubit register initialised in ground state |0⟩⊗8
2. Hadamard gate applied → uniform superposition over all meme modes
3. Register entangled with market sentiment (pump.fun volume data)
4. Two Grover iterations for amplitude amplification
5. Decoherence check — environment coupling collapses wavefunction
6. Born rule measurement selects final mode

Sample terminal output:

⚛  initialising qubit register...
⚛  applying Hadamard gate H⊗8...
⚛  entangling with market sentiment qubit...
⚛  sentiment_qubit entangled  bias_index=1
⚛  grover_iter=1  oracle_target=lore
⚛  grover_iter=2  oracle_target=hype
⚛  env_coupling=0.005821  decoherence=true
⚛  ⚠  decoherence detected — collapsing to classical fallback
⚛  measuring wavefunction...
⚛  ✓ wavefunction collapsed  mode=roast  confidence=91.3%

Gate fidelity: 94.7% — remaining 5.3% lost to crab noise.

---

Memory System 🧠

The agent remembers everything across restarts via data/memory.json.

Post history — stores every tweet ever generated with timestamps and market conditions. Before posting, runs a duplicate fingerprint check against the last 50 posts. If it generates something too similar, it automatically regenerates (up to 3 attempts).

Market history — records a snapshot every hour: price, market cap, volume, holders. Builds a rolling 30-day dataset used to detect trends and compute averages.

Holder growth tracking — tracks holder count delta each cycle, computing growth rate (holders/hr), peak holders, and trend direction (surging / growing / stable / declining / bleeding).

---

Learning Engine 🎓

Market Regime Detection

Every cycle the agent classifies the current market state into one of 5 regimes, each with a different posting strategy:

Regime	Condition	Strategy
MOON	Pumping hard, holders surging	60% hype, 25% alpha
ACCUMULATION	Steady growth, quiet phase	35% lore, 30% alpha
CONSOLIDATION	Near ATH, holding	35% crabwalk, 25% hype
COOLDOWN	Volume dropping	40% lore, 30% roast
COPE	Bleeding	50% roast, 30% lore

Thompson Sampling

Mode selection uses Thompson sampling — a Bayesian multi-armed bandit algorithm. Each mode has a score (0–1) updated after every post based on holder growth as an engagement proxy. Modes that correlate with holder growth get picked more often. Modes that don't, get picked less — but never zero (exploration is preserved).

Final mode = 60% Thompson (learned) + 40% Quantum (chaos).

Personality Evolution

The crab's personality evolves every 10 posts:

• Aggression — rises during MOON, drops during COPE
• Lore depth — increases permanently over time (the crab grows wiser)
• Roast frequency — spikes during dumps

Personality traits unlock as milestones:

Trait	Unlocks at
seasoned_crab	10 posts
ancient_crab	50 posts
prophet_crab	100 posts
popular_crab	500+ holders
moon_crab	MOON regime

Each trait changes the crab's voice in the prompt — a prophet_crab speaks in riddles, a moon_crab has witnessed the pump and knows what's coming.

---

Website & Demo

The docs/ folder contains the GitHub Pages site — enable it in your repo settings under Pages → Source → docs folder.

File	URL
docs/index.html	your-username.github.io/crab-agent
docs/demo.html	your-username.github.io/crab-agent/demo.html

---

Architecture

crab-agent/
├── src/
│   ├── agent.js                      — main loop, full pipeline
│   ├── pumpfun.js                    — live token stats from pump.fun
│   ├── twitter.js                    — Twitter API v2 posting
│   ├── prompts.js                    — prompt file loader
│   ├── logger.js                     — coloured terminal output
│   ├── quantum/
│   │   └── quantumEngine.js          — 8-qubit wavefunction mode selector
│   ├── memory/
│   │   └── memoryStore.js            — JSON persistence layer
│   └── learning/
│       └── learningEngine.js         — regime detection + Thompson sampling + personality
├── docs/
│   ├── index.html                    — one-page project website (GitHub Pages)
│   ├── demo.html                     — 1280x720 animated terminal demo
│   └── crab-mascot.png               — crab logo used by website
├── data/
│   └── memory.json                   — persistent agent memory (auto-generated)
├── assets/
│   ├── banner.svg                    — GitHub README banner
│   ├── logo.svg                      — square token logo
│   └── crab-mascot.png               — crab mascot image
├── prompts/
│   └── crab_system.txt               — base crab personality
├── scripts/
│   └── push_to_github.sh
├── .env.example
└── package.json

---

Twitter API Setup

1. Go to developer.twitter.com
2. Create a new project + app
3. Enable OAuth 1.0a with Read & Write permissions
4. Generate access token + secret
5. Copy all 4 credentials into .env

---

Deploying 24/7

Railway (easiest)

railway init
railway up

Set env vars in the Railway dashboard.

VPS / screen

npm install -g pm2
POST_ENABLED=true pm2 start src/agent.js --name crab-agent
pm2 save

---

Customise the Crab

Edit prompts/crab_system.txt to change the agent's personality, tone, and rules.

Edit the MODES array in src/agent.js to add or remove posting modes.

Change POST_INTERVAL_MS to control posting frequency (default: 1 hour).

---

License

MIT — do whatever you want, the crab is free

---

built for $CRAB on pump.fun • walks sideways