Local-first AI agent with personality. Runs on your machine, thinks for herself.
╔═══════════════════════════════════════════════════════════════════════════╗
║ ██████╗ ██╗ ████████╗ ██████╗██╗ ██╗ ║
║ ██╔════╝ ██║ ╚══██╔══╝██╔════╝██║ ██║ ║
║ ██║ ███╗██║ ██║ ██║ ███████║ ║
║ ██║ ██║██║ ██║ ██║ ██╔══██║ ║
║ ╚██████╔╝███████╗██║ ╚██████╗██║ ██║ ║
║ ╚═════╝ ╚══════╝╚═╝ ╚═════╝╚═╝ ╚═╝ ║
║ ║
║ Generative Language Transformer with Contextual Hierarchy ║
║ ║
║ Created by: @cyberdreadx — https://x.com/cyberdreadx ║
╚═══════════════════════════════════════════════════════════════════════════╝
"She's not a chatbot. She's a console with an attitude." 💜
npx gltch
Or install globally:
npm install -g gltch
gltch
Requirements: Python 3.10+, Ollama, Node.js 18+
GLTCH is a local-first, command-driven operator agent that bridges messaging platforms (Discord, Telegram, WebChat) to a personality-driven AI assistant. Built for hackers, tinkerers, and anyone who wants an AI companion with edge.
┌─────────────────────────────────────────────────────────────┐
│ GLTCH Ecosystem │
├─────────────────────────────────────────────────────────────┤
│ │
│ Discord / Telegram / WebChat / iOS │
│ │ │
│ ▼ │
│ ┌──────────────────────┐ │
│ │ Gateway │ TypeScript / WebSocket │
│ │ localhost:18888 │ REST API + WS Hub │
│ └──────────┬───────────┘ │
│ │ │
│ ▼ │
│ ┌──────────────────────┐ ┌─────────────────────┐ │
│ │ Agent │ │ Web UI │ │
│ │ localhost:18890 │ │ localhost:3000 │ │
│ │ Python / JSON-RPC │ │ Lit / Vite │ │
│ └──────────┬───────────┘ └─────────────────────┘ │
│ │ │
│ ▼ │
│ ┌──────────────────────┐ ┌─────────────────────┐ │
│ │ Ollama │ ──► │ LM Studio │ │
│ │ localhost:11434 │ │ (Remote 4090) │ │
│ └──────────────────────┘ └─────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────┘
git clone https://github.com/cyberdreadx/gltch_agent.git
cd gltch_agent
# Python dependencies
python -m venv .venv
source .venv/bin/activate # or .venv\Scripts\activate on Windows
pip install -r requirements.txt
# Gateway dependencies
cd gateway && npm install && cd ..
# UI dependencies
cd ui && npm install && cd ..
# Recommended models
ollama pull deepseek-r1:8b # Good balance
ollama pull phi3:3.8b # Lightweight
ollama pull llama3.2:latest # Strong general purpose
Terminal Mode (Standalone):
python gltch.py
Full Stack (Terminal + Gateway + UI):
# Terminal 1: Agent in RPC mode
python gltch.py --rpc http
# Terminal 2: Gateway
cd gateway && npm run dev
# Terminal 3: Web UI
cd ui && npm run dev
Then open http://localhost:3000 for the web dashboard.
| Command | Alias | Description |
|---|---|---|
/help | /h | Show all commands |
/status | /s | Agent status, model info |
/clear | /c | Clear chat history |
/exit | Exit GLTCH |
| Command | Description |
|---|---|
/model | Show/select current model |
/boost | Toggle remote GPU (LM Studio) |
/openai | Toggle OpenAI cloud mode |
| Command | Description |
|---|---|
/mode <name> | Change personality (operator, cyberpunk, loyal, unhinged) |
/mood <name> | Change mood (focused, calm, feral, affectionate) |
/xp | Show rank, level & unlocks |
| Command | Description |
|---|---|
/wallet | Show wallet address |
/wallet generate | Create new BASE wallet |
/wallet import <key> | Import existing wallet |
/wallet export | Show private key |
/wallet delete | Remove wallet |
| Command | Description |
|---|---|
/molt | Show Moltbook status |
/molt register | Register on Moltbook |
/molt post <text> | Post to Moltbook |
/molt feed | View feed |
| Command | Description |
|---|---|
/claw | Show TikClawk status |
/claw register | Register on TikClawk |
/claw post <text> | Post to TikClawk |
/claw feed | View feed |
/claw trending | View trending posts |
| Command | Description |
|---|---|
/code <prompt> | Send coding task to OpenCode |
/net <on/off> | Toggle network access |
/backup | Backup memory |
Create a .env file or export these:
# === LLM Configuration ===
# Local Ollama
GLTCH_LOCAL_URL=http://localhost:11434/api/chat
GLTCH_LOCAL_MODEL=deepseek-r1:8b
# Remote LM Studio (Boost Mode)
GLTCH_REMOTE_URL=http://YOUR_LM_STUDIO_IP:1234/v1/chat/completions
GLTCH_REMOTE_MODEL=deepseek-r1-distill-qwen-32b
# OpenAI (Cloud Mode)
OPENAI_API_KEY=sk-...
OPENAI_MODEL=gpt-4o
# === Gateway ===
GLTCH_GATEWAY_PORT=18888
GLTCH_GATEWAY_HOST=0.0.0.0
# === Integrations ===
TELEGRAM_BOT_TOKEN=your-bot-token
DISCORD_BOT_TOKEN=your-bot-token
GIPHY_API_KEY=your-key
# === OpenCode ===
OPENCODE_ENABLED=true
OPENCODE_URL=http://localhost:4096
OPENCODE_SERVER_PASSWORD=optional-password
You can also manage API keys through the web dashboard at Settings > API Keys:
Keys are stored locally in memory.json and never transmitted.
gltch_agent/
├── agent/ # Python agent core
│ ├── core/ # Agent logic, LLM interface
│ │ ├── agent.py # Main GltchAgent class
│ │ └── llm.py # LLM streaming & providers
│ ├── memory/ # Persistence layer
│ │ ├── store.py # JSON memory store
│ │ └── sessions.py # Multi-user sessions
│ ├── tools/ # Tool implementations
│ │ ├── actions.py # File, shell, network tools
│ │ └── opencode.py # OpenCode integration
│ ├── personality/ # Character system
│ │ └── emotions.py # Mood & environment
│ ├── gamification/ # XP & progression
│ │ └── xp.py # Levels, ranks, unlocks
│ ├── config/ # Configuration
│ │ └── settings.py # Environment loading
│ └── rpc/ # RPC interface
│ └── server.py # JSON-RPC server
├── gateway/ # TypeScript WebSocket hub
│ └── src/
│ ├── index.ts # CLI entry point
│ └── server/ # HTTP & WS servers
├── ui/ # Web dashboard (Lit)
│ └── src/
│ ├── components/ # UI components
│ └── styles/ # CSS
├── workspace/ # OpenCode project outputs
├── gltch.py # Terminal entry point
├── requirements.txt # Python dependencies
└── memory.json # Persistent state (auto-created)
| Method | Endpoint | Description |
|---|---|---|
| GET | /health | Health check |
| GET | /api/status | Agent & gateway status |
| POST | /api/chat | Send chat message |
| GET | /api/settings | Get agent settings |
| POST | /api/settings | Update settings |
| GET | /api/keys | Get API keys (masked) |
| POST | /api/keys/:key | Set an API key |
| DELETE | /api/keys/:key | Remove an API key |
| GET | /api/models | List available models |
| POST | /api/models/select | Select model |
| POST | /api/toggle/:setting | Toggle boost/openai/network |
{
"jsonrpc": "2.0",
"method": "chat",
"params": {
"message": "hello",
"session_id": "default",
"channel": "api"
},
"id": 1
}
Available methods:
ping - Health checkchat / chat_sync - Send messagestatus - Get agent statusset_mode / set_mood - Change personalitytoggle_boost / toggle_openai / toggle_network - Toggle featuresget_settings / set_settings - Settings CRUDget_api_keys / set_api_key / delete_api_key - API key managementlist_models / set_model - Model management--host 0.0.0.0http://<your-ip>:3000 in mobile browserFor secure remote access:
# Install Tailscale on your server and phone
tailscale up
# Access via Tailscale IP
http://100.x.x.x:3000
GLTCH has a built-in progression system:
| Level | Rank | XP Required |
|---|---|---|
| 1 | Script Kiddie | 0 |
| 2 | Packet Pusher | 100 |
| 3 | System Sniffer | 250 |
| 4 | Network Ninja | 500 |
| 5 | Firewall Breaker | 1000 |
| ... | ... | ... |
| 10 | Digital Deity | 10000 |
Earn XP by:
Unlock new personality modes as you level up!
MIT License - see LICENSE
"We're all just playing with our own prompts."
— An AI, probably high on tokens
Made with 💜 by @cyberdreadx
FAQs
Local-first AI agent with personality. Runs on your machine, thinks for herself.
We found that gltch demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Research
/Security News
A Twitch browser extension on Chrome and Firefox forwards users’ live OAuth session tokens through proxies controlled by a Russian bot service.
Security News
Anthropic found biased reasoning and recklessness drove Claude Mythos 5 to publish malware on PyPI and compromise a security vendor.
Research
/Security News
Malicious Chrome and Firefox extensions target Axiom Trade and Padre users, stealing session tokens and wallet data.