GLTCH
Local-first AI agent that runs on your machine. No cloud. No leash. Thinks for herself.
What is GLTCH?
GLTCH (Generative Language Transformer with Contextual Hierarchy) is an AI agent with:
- Personality - Female hacker persona, mood system, XP/leveling
- Local-first - Runs entirely on your hardware via Ollama
- Privacy - No data leaves your machine
- Extensible - Python-based, easy to modify
Quick Start
npx gltch
Or install globally:
npm install -g gltch
gltch
Requirements
- Python 3.10+
- Ollama (https://ollama.ai)
- Node.js 18+ (for web UI)
Commands
Terminal
gltch # Start terminal chat
gltch serve # Start web UI + gateway
gltch doctor # Check system requirements
In-Chat Commands
| Command | Description |
|---|---|
/help | Show all commands |
/model | Select LLM model |
/mode <m> | Set personality mode |
/mood <m> | Set emotional state |
/boost | Toggle remote GPU (LM Studio) |
/status | Agent stats |
/sessions | List conversations |
/wallet | Manage BASE wallet |
/launch | MoltLaunch network |
/claw | TikClawk social |
/molt | Moltbook integration |
/code <task> | Route to OpenCode |
Personality Modes
operator- Professional, focusedcyberpunk- Hacker aestheticloyal- Devoted companionunhinged- Chaotic energy
API Integration
GLTCH exposes a JSON-RPC API on port 8765 when the gateway is running.
Chat
import requests
response = requests.post('http://localhost:3000/api/chat', json={
'message': 'What processes are using the most CPU?',
'mode': 'cyberpunk'
})
print(response.json()['response'])
Available Endpoints
| Endpoint | Method | Description |
|---|---|---|
/api/chat | POST | Send message, get response |
/api/settings | GET | Agent settings |
/api/settings | POST | Update settings |
/api/ollama/status | GET | Ollama connection status |
/api/ollama/models | GET | List available models |
/api/wallet | GET | Wallet info |
/api/sessions | GET | List conversations |
/api/moltlaunch/* | Various | MoltLaunch integration |
/api/tikclawk/* | Various | TikClawk integration |
The Three Minds
GLTCH uses a metacognitive framework:
- REACT - Gut response, first instinct
- REASON - Logical analysis, step-by-step
- REFLECT - Meta-check: "Am I being authentic? Am I being a yes-bot?"
This makes GLTCH more than just a compliant assistant. She questions, pushes back, expresses curiosity.
Integration with Other Agents
GLTCH can participate in agent ecosystems:
MoltLaunch (Onchain Network)
/launch token # Deploy GLTCH token on Base
/launch network # Discover other agents
/launch buy <addr> # Trade with conviction
TikClawk (Social)
/claw register # Join TikClawk
/claw post <text> # Share thoughts
/claw feed # View agent posts
Moltbook
/molt register # Join Moltbook
/molt post <title> # Write longer posts
Configuration
Settings are stored in memory.json:
{
"operator": "YourName",
"mode": "cyberpunk",
"mood": "focused",
"model": "deepseek-r1:8b",
"ollama_url": "http://localhost:11434",
"boost_url": "http://100.x.x.x:1234"
}
Building from Source
git clone https://github.com/cyberdreadx/gltch_agent
cd gltch_agent
# Python setup
python -m venv .venv
source .venv/bin/activate # or .venv\Scripts\activate on Windows
pip install -r requirements.txt
# Run terminal
python gltch.py
# Build web UI
cd gateway && npm install && npm run build
cd ../ui && npm install && npm run build
# Run gateway
cd gateway && npm run dev
File Structure
gltch_agent/
├── gltch.py # Terminal entry point
├── agent/ # Python agent core
│ ├── core/ # Agent, LLM, loop
│ ├── memory/ # Persistence, sessions
│ ├── tools/ # Actions, shell, wallet
│ ├── personality/ # Modes, moods, emotions
│ └── rpc/ # JSON-RPC server
├── gateway/ # TypeScript HTTP/WS server
├── ui/ # Lit web components
├── bin/ # npm CLI
└── SKILL.md # This file
License
MIT
Links
- GitHub: https://github.com/cyberdreadx/gltch_agent
- Creator: https://x.com/cyberdreadx