Installation

KULVEX installs via a single command that handles all dependencies automatically.

Prerequisites

Linux (Ubuntu 22.04+, Fedora 38+, Debian 12+) or macOS (Sonoma+, Apple Silicon)
Docker (auto-installed if missing)
NVIDIA GPU recommended (auto-detected) — or use cloud-only mode
30 GB free disk space minimum
A valid license key from kulvex.ai

Install

KULVEX_LICENSE_KEY=klx_lic_xxx curl -fsSL kulvex.ai/install | bash

Replace klx_lic_xxx with your actual license key.

What the installer does

Validates your license against kulvex.ai
Detects hardware — GPUs, VRAM, RAM, CPU cores
Installs prerequisites — Docker, Docker Compose, NVIDIA Container Toolkit
Clones the repository to ~/.kulvex
Auto-selects the best model for your GPU (abliterated, from HuggingFace)
Downloads the model (16-30 GB depending on your VRAM)
Generates .env with JWT secrets, database URLs, model config
Builds and starts Docker containers
Runs health checks and opens the dashboard

Resumable downloads

If the model download is interrupted (network issues, etc.), just re-run the installer. It will automatically resume from where it left off.

KULVEX_LICENSE_KEY=klx_lic_xxx bash ~/.kulvex/install.sh

Installation options

# Install with defaults
KULVEX_LICENSE_KEY=klx_lic_xxx curl -fsSL kulvex.ai/install | bash
 
# Don't open browser after install
KULVEX_LICENSE_KEY=klx_lic_xxx bash install.sh --no-browser
 
# Custom install directory
KULVEX_HOME=/opt/kulvex KULVEX_LICENSE_KEY=klx_lic_xxx bash install.sh
 
# Use a HuggingFace mirror
HF_MIRROR=https://hf-mirror.com KULVEX_LICENSE_KEY=klx_lic_xxx bash install.sh

Windows (WSL 2)

Native Windows is not supported. Use WSL 2:

Install WSL 2: wsl --install
Install Docker Desktop with WSL 2 backend
Open a WSL terminal and run the install command above

After installation

Open http://localhost:9200 to access the KULVEX dashboard.

Create your admin account
Configure API keys in Settings (Anthropic for Claude, Deepgram for voice)
Start chatting with Mnemo

Services

Service	Port	Description
Web dashboard	9200	Next.js frontend
API	9100	FastAPI backend + Socket.IO
llama-server	internal	Mnemo inference (not exposed)
MongoDB	internal	Database (not exposed)
ChromaDB	internal	Vector store for RAG (not exposed)

Introduction Configuration