Generate images. Clone voices. Automate everything. Privately.

Stop renting AI.
Own it.

Chat with AI personalities that remember you. Generate images and video. Clone any voice. Automate your workflows. Search the web privately. All on your own hardware. One purchase. Zero subscriptions. Zero data leaked. Ever.

Download Free Watch Demo
$0/mo After Purchase
0 AI Services Included
0 bytes Data Leaked
1 cmd To Launch Everything
Scroll to explore
The Problem

You're paying $100+/month to train someone else's AI.

Every message you send to ChatGPT, Midjourney, or ElevenLabs becomes their training data. You pay monthly. They own your creativity. Stop when you want — lose everything.

ChatGPT Plus
$20/mo
Midjourney
$30/mo
ElevenLabs
$22/mo
Notion AI
$10/mo
Zapier
$20/mo
Total / Year
$1,224
...and they own your data
SpiritBox Ultimate One Purchase. Own Forever.
0 bytes sent to cloud
What You Get

Everything a cloud AI can do — but it's yours.

spiritbox-ultimate
$ docker compose up
spiritbox-ultimate    Healthy
spiritbox-comfyui    Healthy
spiritbox-voice      Healthy
spiritbox-moltbot    Healthy
ultimate-graph      Healthy
ultimate-db         Healthy
ultimate-cache      Healthy
ultimate-search     Healthy
ultimate-n8n        Healthy
10/10 containers ready. VRAM Master online.
SpiritBox Ultimate v1.01 "VRAM Relay"
$ _

One Command. You're Live.

Type docker compose up and walk away. Ten AI services launch, connect, and health-check themselves. Come back to a fully operational AI assistant. No setup guides. No headaches.

AI That Knows You

5 AI personalities with different skills. Ask Nova to design your logo. Ask Echo to write your code. Ask Sage to research your competitors. They remember your projects and speak in cloned voices.

Any GGUF ModelSelf-Hosted EnginePersonality Engine

Create Anything Visual

Type what you want. Get images, videos, and animations back. Logos, product photos, social media content, art — generated on your GPU. You own every pixel. No licensing fees.

ComfyUISDXL / FluxVRAM Relay

Clone Any Voice

Upload a 30-second audio clip. Get a cloned voice back. Your AI assistants speak in any voice you choose — your own, a character, a brand persona. Entirely local. Entirely private.

Piper TTSWhisper STTVoice Lab

An AI That Remembers

Your AI builds a personal knowledge graph from your conversations. It remembers your projects, preferences, and context — but only facts you explicitly approve. Your memory, your control.

Neo4jpgvectorConsent-Gated

Automate Your Life

"Back up my data every Sunday." "Summarize my emails." "Generate reports weekly." Built-in n8n automation does the repetitive work. Trigger from chat, Telegram, or set a schedule.

n8nWebhooksCredential Isolation

Search Without Being Watched

Google tracks every search. SpiritBox doesn't. Built-in private search engine returns real results without profiling you, logging your queries, or selling your search history.

SearXNGZero TrackingAir-Locked

Run Code Safely

Ask your AI to write a script and it executes in a sealed sandbox. No internet access, no access to your passwords. The code runs, you get results, nothing leaks.

MoltbotNetwork IsolatedSafe Execution

Your AI, Everywhere

Message your home AI from your phone via Telegram. Send a photo for analysis. Ask it to generate an image. Trigger automations on the go. Whitelist-protected so only you get in.

TelegramCloudflare TunnelWhitelist
Architecture

10 containers. 3 security zones. Zero trust.

Every service runs in its own Docker container with principle-of-least-privilege. The Brain reasons. The Hands execute. The Vault holds secrets. They never cross.

The Brain (Reasoning)
SpiritBox App :6781
LLM Engine :8000
The Hands (Execution)
ComfyUI:8188
Piper TTS:8778
Moltbot:18789
n8n:5678
SearXNG:8080
The Vault (Storage)
Neo4j:7687
PostgreSQLInternal
RedisRead-Only
no-new-privileges on all 9 services
PostgreSQL has zero host ports
Redis is read-only with tmpfs
Zero outbound telemetry confirmed
Patent Pending

VRAM Relay™

The only system that runs a 14B-parameter LLM, Stable Diffusion XL, Whisper, and Piper TTS on a single consumer GPU without out-of-memory crashes.

  • Automatically unloads LLM before image generation
  • LLM Engine auto-reloads on next chat request
  • Cooldown timing prevents VRAM fragmentation
  • Works on RTX 3090, 4090, and 5090
GPU VRAM (24 GB)
LLM 18GB
Free 6GB
User: "Generate an image"
ComfyUI 14GB
Free 10GB
VRAM Relay unloads LLM → ComfyUI gets full GPU → LLM auto-reloads on next chat
Landscape

Nobody else does what we do.

Single-purpose tools solve one problem. Enterprise platforms cost $10K+. SpiritBox fills the gap.

Capability SpiritBox Ultimate Ollama Jan AI Open WebUI PrivateGPT
LLM Chat
Image / Video Generation
Voice Cloning & TTS
Knowledge Graph Memory
Workflow Automation (n8n)
Sandboxed Code Execution
Private Web Search
Telegram / Remote Access
VRAM Orchestration
AI Persona & Voice Engine
Glass Box Security

Privacy isn't a feature. It's the architecture.

BYO-Keys Vault

API keys encrypted with Fernet AES-128-CBC. The AI decides what to do; a separate script executes without showing the key to the LLM.

Prompt Injection Defense

Aegis Sanitizer with 12+ injection patterns. Every RAG input is scrubbed. Every web search result is cleaned before reaching the LLM.

Container Isolation

no-new-privileges + cap_drop: ALL on services. Moltbot has zero network access. PostgreSQL has zero host ports.

Fully Auditable

Bandit static analysis: 0 high findings. Trivy container scan: 301 CVEs (22% reduction), zero exploitable in local threat model. Run scans yourself anytime.

See It In Action

One prompt. Five superpowers.

Click each tab to see what happens when you type a single message into SpiritBox. Everything below runs locally on your hardware — no cloud, no subscription, no data leaves your machine.

“Help me plan my product launch”
Atlas Project Strategist

Spirit Guides are specialized AI personalities, each with unique expertise, voice, and a custom system prompt honed for their domain. Switch guides mid-conversation — or let SpiritBox auto-route to the right expert.

Nova Creative Director — Images, video, design concepts. Drives ComfyUI workflows for text-to-image, img-to-video, and more.
Sage Research Analyst — Deep dives, competitive analysis, web searches via SearXNG. Generates charts and Mermaid diagrams.
Atlas Project Strategist — Phase-based planning, milestones, Kanban boards. Triggers n8n automations for task tracking.
Echo Tech Specialist — Code generation, debugging, system diagnostics. Executes scripts in the sandboxed Moltbot container.
Ember Storyteller — Creative writing, marketing copy, social media. Adapts tone, length, and style to your audience.
Create Your Own Custom system prompt, voice, persona, tool access, and specialized behavior. Unlimited guides. Your rules.
Each guide speaks with a unique cloned voice via Piper TTS · All conversations stored locally
“Create a cyberpunk cityscape at sunset”
Nova Creative Director

SpiritBox doesn't just chat — it acts. Your Spirit Guide detects what you need and triggers the right tool automatically. No manual setup, no API keys in the chat window.

ComfyUI — Image & Video Generation VRAM Relay auto-unloads the LLM, fires your GPU workflow, returns the image inline. Text-to-image, image-to-video, img2img — any ComfyUI workflow you install.
GPU
SearXNG — Private Web Search AI-summarized search results from 70+ engines. No tracking, no ads, no search history sent anywhere.
Web
Moltbot — Sandboxed Code Execution Python, shell scripts, system diagnostics. Runs in an isolated container — can't touch your host filesystem.
Code
n8n — Workflow Automation Trigger multi-step automations: email, Telegram, webhooks, file ops. The AI plans, n8n executes.
Auto
Piper TTS + Whisper STT Your AI speaks back in a cloned voice. You speak to it via Whisper. Full voice loop — including Bluetooth headset support.
Voice
AI-generated cyberpunk cityscape
Generated on your GPU in 4.2s · You own this image · 0 bytes sent to cloud
“Research my competitors, build a launch plan, create a logo, and draft the announcement email”
SpiritBox Orchestrator

This is the killer feature. One prompt triggers a chain of actions across multiple tools and guides. SpiritBox reads your request, breaks it into steps, and executes them in sequence — automatically.

1
Sage researches competitors SearXNG
Searches 70+ engines, summarizes top 5 competitors, identifies market gaps
2
Atlas builds launch plan Task Board
6-week phased plan with milestones, dependencies, and deliverables
3
Nova generates logo concepts ComfyUI
VRAM Relay swaps LLM for Stable Diffusion, generates 4 logo variations
4
Ember drafts announcement email Writing
Uses research context + brand voice to write conversion-optimized copy
4 guides · 3 tools · 1 prompt · Zero human intervention between steps
“Every Monday at 9am, check my competitors' pricing and send me a Telegram summary”
Atlas via n8n Automation

SpiritBox isn't just reactive — it runs while you sleep. n8n workflows trigger on a schedule, invoke Spirit Guides, use tools, and deliver results to you however you choose.

Cron Trigger Every Monday 9:00 AM
Active
SearXNG scans competitor sites Price tables, new products, social mentions
Sage analyzes changes Delta report vs. last week, risk flags, opportunities
Telegram delivers the briefing Formatted summary on your phone before your first coffee
Runs on your hardware 24/7 · No API costs · Your competitive intel never touches a third-party server
“Build me a complete marketing campaign for my new product”
Dashboard Multi-Guide Collaboration

The Dashboard is your mission control. Multiple Spirit Guides work on your request in parallel — each contributing their specialty, all visible in real-time on a Kanban board with live logs.

In Progress
Sage
Market Research Analyzing competitor positioning, audience demographics, pricing strategy...
Nova
Visual Assets Generating hero banner, social media templates, product mockups via ComfyUI...
Queued
Ember
Email Sequence Waiting for Sage's research to write audience-targeted copy...
Atlas
Launch Timeline Will build Gantt chart from all deliverables once assets are ready...
Complete
Echo
Landing Page Scaffold HTML/CSS template generated via Moltbot, files saved to workspace.
5 guides · 4 tools · 1 Kanban board · Live logs · Final report auto-generated
Try the full interactive demo
Voice Lab

Clone any voice. Locally.

Upload a 30-second audio clip. SpiritBox trains a custom voice model on your GPU using Piper TTS and Whisper. Your voice samples never leave your machine.

1
Upload audio
WAV, MP3, FLAC — 30 seconds minimum
2
Whisper transcribes
Auto-sliced and aligned for training
3
Piper fine-tunes
GPU-accelerated transfer learning
4
Your voice model
Export .onnx for instant TTS
Download Voice Lab Docker Standalone Docker app · Works without SpiritBox Ultimate
SpiritBox Voice Lab
Training — Epoch 1,560 / 2,000 78%
my-voice.onnx 18.4 MB
my-voice.onnx.json 1.2 KB
Get Started

Start free. Upgrade when you're ready.

Download the free version and start chatting with local AI today. When you want the full arsenal, upgrade to Ultimate.

SpiritBox Free

$0
Forever free
  • Chat with local LLMs
  • Voice conversations (Piper TTS + Whisper STT)
  • Spirit Guide personalities
  • Conversation history
  • Self-hosted LLM Engine — runs any GGUF model locally
  • Voice Lab add-on — Clone voices for free
Download Free
Most Popular

SpiritBox Ultimate

$0
One-time purchase · Own forever
  • Everything in Free, plus:
  • Image & video generation (ComfyUI)
  • Voice cloning (built-in Voice Lab)
  • Knowledge graph memory (Neo4j)
  • Private web search (SearXNG)
  • Workflow automation (n8n)
  • Sandboxed code execution (Moltbot)
  • Telegram remote access
  • VRAM Relay™ (patent pending)
  • 10 Docker services, one command launch
Buy Ultimate — Stripe Checkout