SpiritBox Creative Guide Actual interface capture
Develops creative directions, prompts, concepts, and visual approaches.
SPIRITBOX ULTIMATE v1.12 / WINDOWS
The full local AI workstation.
SpiritBox Ultimate is a nine-service Windows local-AI workstation for typed chat, Piper voice output, documents and local recall, private search, n8n automation, user-managed ComfyUI integration, and sequential VRAM Relay coordination.
A nine-service local AI workstation for documents, automation, private search, voice, and GPU-heavy creative workflows. Connect your user-supplied ComfyUI endpoint while VRAM Relay coordinates heavy workloads sequentially. The optional request-only trainer is not one of the nine default services.
$199USD ONE-TIME
ACTUAL v1.12 INTERFACE
The Subscription Trap
Cloud AI adds recurring bills and outside data paths.
Subscription tools can be useful, but they keep the account, billing, and data path outside your workstation. SpiritBox gives you a local-first option you can inspect and run on your own machine.
ChatGPT Plus
$20/mo
Midjourney
$30/mo
ElevenLabs
$22/mo
Notion AI
$10/mo
Zapier Starter
$20/mo
Combined Cloud Stack
$102/mo
$1,224 per year at these plan prices
System Architecture
9 default Docker services. 3 isolated zones.
The Brain reasons, Hands and Voice execute, and the Vault stores data. Nine services start in the default Compose fleet; the Voice Lab trainer is an optional profile, and ComfyUI connects through the endpoint you configure.
The Brain
REASONING & ROUTING
SpiritBox Core App
:6781
Local LLM Engine (GGUF)
:8000
VRAM Relay Manager
Internal
Hands and Voice
CREATION, SPEECH & EXECUTION
Configured ComfyUI Endpoint
Host / LAN
Piper TTS service
:8778
Moltbot Sandboxed Exec
:18789
n8n Automation Engine
:5678
SearXNG Private Search
:8080
The Vault
KNOWLEDGE & MEMORY
Neo4j Knowledge Graph
:7687
PostgreSQL + pgvector
Zero Host Ports
Redis (tmpfs Read-Only)
RAM Only
Fernet AES-128 Key Vault
Encrypted
no-new-privileges enforced across all containers PostgreSQL has zero exposed host ports
Moltbot code executor is network-isolated
Zero outbound telemetry or telemetry pingback
Live Simulation
One prompt. Five superpowers.
Click through each tab below to see how SpiritBox turns a single prompt into actions across guides, tools, workflows, and document knowledge on your hardware.
“Help me plan my product launch strategy and visual identity”
Choose the guide built for the job.
Spirit Guides use distinct system instructions and tool access. Voices are selected separately, so any installed voice can be assigned to the guide you choose.
SpiritBox Technical Guide Actual interface capture
Helps diagnose software, explain systems, and work through technical problems.
Market Research Analyst Assistant
Structures market questions, competitor reviews, and evidence summaries.
n8n Architect
Plans triggers, actions, approval points, and recovery paths for n8n workflows.
Product Description Writer
Turns product facts into clear descriptions without losing important constraints.
Sales Pitch Guide
Builds practical sales pitches around the buyer, proof, objections, and next action.
“Render a photorealistic titanium hardware mockup at 4K resolution”
SpiritBox Creative Guide Triggering a ComfyUI workflow
SpiritBox doesn't just output text — it orchestrates your local tools. When the Creative Guide needs to generate visual assets, VRAM Relay can unload the language model, execute a compatible ComfyUI workflow on the GPU, and reload the LLM. The workloads run sequentially and still have to fit the selected hardware.
“Research competitor pricing, generate a marketing banner, and write our launch email”
SpiritBox Orchestrator Multi-Step Autonomous Pipeline
One single prompt coordinates multiple guides and tools sequentially without human intervention between steps.
1
Market Research Analyst Assistant: Competitor Research
Structures the search, reviews the gathered results, and identifies useful price and positioning differences.
2
Digital Foreman: Structured Positioning Matrix
Organizes the findings against the supplied product facts and drafts a practical differentiator matrix.
3
SpiritBox Creative Guide: Hero Marketing Banner (ComfyUI + VRAM Relay)
Dynamically allocates GPU memory to render 3 brand banners matching the target color palette.
4
Email Campaign Drafter Assistant: Launch Email
Drafts the complete announcement email, incorporating the research context and image attachments.
“Every morning at 7 AM, scan industry news, summarize key shifts, and send to my Telegram”
n8n Workflow Engine 24/7 Background Automation
SpiritBox runs while you sleep. Built-in n8n workflow triggers wake the system on cron schedules, execute deep research, and securely push briefings to your phone via encrypted Telegram bots.
[07:00:00] Cron trigger fired -> "Daily Intelligence Briefing"
[07:00:02] SearXNG queried 14 news feeds and arXiv papers
[07:00:08] Market Research Analyst summarized 3 relevant releases
[07:00:10] Telegram Bot dispatched encrypted Markdown brief to User ID #8849201
Telegram: "Good morning Ryan. 3 new local model weights were published overnight..."
“Execute our full Q3 marketing and technical release campaign”
TaskBoard Mission Control Parallel Multi-Agent Execution
Use the TaskBoard to review and route work through configured guides and workflows. Displayed task movement is a scripted product walkthrough, not proof of simultaneous model execution.
In Progress
2
SpiritBox Creative Guide — Visual Asset Generation
Rendering 4K product hero shots in ComfyUI...
Digital Foreman — Gantt Alignment
Building phased work breakdown structure...
Queued
2
Product Description Writer — Product Copy
Waiting for technical spec sign-off...
SpiritBox Technical Guide — Docker Health Verification
Scheduled after asset bundling completes...
Complete
1
Market Research Analyst Assistant — Competitor Pricing Audit
Report compiled and saved to the project workspace.
Patent Pending
VRAM Relay™ GPU Coordination
Run large 14B parameter language models, Stable Diffusion XL / Flux, Whisper, and Piper TTS on a single compatible NVIDIA GPU by coordinating which configured workload holds VRAM at a given time.
- Fewer memory conflicts: Coordinates unloading the LLM before a configured ComfyUI workflow requests the GPU.
- Instant Auto-Reload: LLM seamlessly reloads back into VRAM on the very next chat prompt.
- Fragmentation Guard: Cooldown timers and memory defragmentation keep long sessions stable.
GPU Memory Allocation (24 GB RTX 4090)
Dynamic
User Prompt: "Generate high-res product mockup in ComfyUI"
VRAM Relay coordinates sequential memory handoffs between configured workloads. It does not make every model or workflow fit; results depend on the GPU, models, workflows, and active configuration.
Knowledge Base
Bring your own documents into the conversation.
Ultimate can extract text from supported files, index it with pgvector, and help you find relevant passages by meaning. You review the matches before adding them to the chat.
Upload
Add PDF, DOCX, TXT, CSV, XLSX, JSON, HTML, Markdown, or RTF files. Originals remain in your workspace.
Recall by meaning
Use >>recall <topic> to search the local pgvector index for relevant passages, not just matching words.
Review and use
Select the passages you want before SpiritBox adds them to the active conversation context.
Real Application UI
Built for power users. Engineered with precision.
No mockups or concept renders. This is the real SpiritBox Ultimate interface running on local hardware. Click any card to inspect full-resolution details.
Command Center Dashboard
The v1.12 local command center with direct access to chat, workflows, the Knowledge Base, voices, and system settings.
Spirit Guides Roster
Manage, customize, and export AI guides with individual system instructions and Piper voice models.
Voice Studio & Voice Lab
Manage the included compatible Piper TTS voices and adjust playback controls. The separate Voice Lab trainer remains request-only and is not part of the qualified public package.
ComfyUI Rituals Registry
Registered pipelines for Text-to-Image, Image-to-Video, and LTX models with VRAM tracking per workflow.
v1.12 Fleet Health
Live status for core services and optional integrations, with local hardware and connection boundaries shown in the app.
Built in the open.
Follow the experiments, failures, road trips, and working releases behind SpiritBox. The dev journey keeps the long-form story separate from the product pitch.
Comparison Matrix
One local workstation, with explicit boundaries.
Single-purpose tools only solve chat. Cloud platforms trap your data. SpiritBox Ultimate delivers a unified sovereign workstation.
| Capability | SpiritBox Ultimate | Ollama | Jan AI | Open WebUI | PrivateGPT |
|---|---|---|---|---|---|
| Local LLM Chat (GGUF) | |||||
| Image & Video Gen (ComfyUI) | |||||
| Piper TTS output | |||||
| VRAM Relay GPU Coordination | (Patent Pending) | ||||
| Document Knowledge Base with pgvector recall | |||||
| Workflow Automation (n8n) | |||||
| Sandboxed Code Execution (Moltbot) | |||||
| Private Search Engine (SearXNG) | |||||
| Remote Telegram Whitelist Access | |||||
| Zero Recurring Subscriptions |
Built to be adapted.
Ultimate is designed for ownership and adaptability: readable local source, customizable Spirit Guides, workflows, and interface, plus a documented brand profile, brand assets folder, and rebrand tooling. Change product text, colours, icons, and terminology without rewriting the backend. Commercial redistribution or client deployment requires the appropriate licence.
Transparent Pricing
Start free. Upgrade when you're ready.
Regular prices are $49 USD for Pro and $199 USD for Ultimate. Use EARLY50 at launch for 50% off either paid SpiritBox edition.
Free Entry
SpiritBox Basic
$0 / forever
The free entry point for private local LLM chat.
- Local LLM Chat (GGUF models)
- Standard Spirit Guides
- Local Conversation History
- Windows Default System Voice
- No Voice Lab Support (Requires Pro/Ultimate)
- Local chat with no required SpiritBox cloud account
Core Paid Edition
SpiritBox Pro
$49 / one-time
Regular $49 USD one-time$24.50 with EARLY50Ends September 30
The focused paid app for natural local voice replies, custom guides, and conversation history. Already chosen by early customers.
- Everything in Basic, plus:
- Piper TTS Engine (AI Chat Talks Back!)
- Typed input with opt-in Piper TTS output
- Local Conversation History
- Custom Spirit Guide Persona Studio
- Lightweight App (No Complex Docker Overhead)
- One-time license for the Pro Windows package
9 Core Services + Optional Trainer
SpiritBox Ultimate
$199 / one-time
Regular $199 USD one-time$99.50 with EARLY50Ends September 30
The full sovereign workstation for creators, developers & automation engineers.
- Everything in Pro, plus:
- 9-Service Docker Fleet + optional Voice Lab trainer
- ComfyUI Studio (Image & Video Gen)
- VRAM Relay™ (Patent Pending GPU Swap)
- Document Knowledge Base + pgvector Semantic Recall
- n8n Automation Engine (24/7 Workflows)
- SearXNG Private Search & Moltbot Sandbox
- Remote Telegram Bot Whitelist
- White-label branding profile and rebrand tooling
Voice Lab remains request-only. Its standalone source ZIP is not a versioned or qualified customer package and is not a public Gumroad product. Ultimate includes an optional trainer profile, not a separate public download. Read the access boundary.
Got Questions?
Frequently Asked Questions
What stays local, and what can connect outside?
Core models, conversations, search indexes, and project data are designed to run on hardware you control. Model downloads, updates, web search, Telegram, ComfyUI on another host, and any API you enable can communicate outside the machine. The Settings and Privacy pages name those boundaries.
What are the recommended hardware system requirements?
Qualified package: Windows 10/11 with Docker Desktop and WSL2. Ultimate's practical baseline is 32GB system RAM. A 12GB NVIDIA GPU is the unqualified entry tier for smaller quantized models and lighter workflows; 24GB is the proven best-fit tier for the current roughly 19GB Qwen3-14B-Q8 configuration, larger models, and demanding media. Review the requirements page before buying.
What does VRAM Relay™ do on one GPU?
VRAM Relay is our patent-pending memory coordination technology. It runs configured LLM and ComfyUI workloads sequentially, unloading and reloading the LLM around compatible media work. It does not make every model or workflow fit, and results depend on the GPU, models, context, and media pipeline.
How does the one-time purchase work?
Gumroad delivers the Windows ZIP and checksum after purchase. The SpiritBox license is a one-time purchase, not a subscription. You are still responsible for your own hardware and for any optional third-party models, APIs, storage, or network services you choose to use.
Can I access SpiritBox from my phone or other devices?
SpiritBox includes LAN access settings for opening the dashboard from another device on your local network. Optional Telegram access uses a configured user-ID allowlist, but it creates an external network and data path with its own account, transport, and configuration boundaries.