Tabby Tavern Stack — A Complete Private AI Platform

THE PROBLEM

Your team pays for code assistants, chat subscriptions, image generation APIs, transcription services, workflow tools, and vector databases — separately, monthly, forever. Every prompt, image, and voice clip leaves your network. You're renting 12 different SaaS products when you could own one stack that does all of it.

THE SOLUTION — THE TABBY TAVERN STACK

A complete, containerized AI platform installed on hardware you control. One docker compose up brings up 21 services across text generation, image synthesis, audio transcription, code intelligence, workflow automation, private search, vector storage, identity management, and remote access — all inside your network. Published and documented on Hugging Face: huggingface.co/jpanasuk/tabby-tavern-stack.

Text & LLM Inference

TabbyAPI — EXL3-quantized models, 40+ tok/s
Ollama — secondary inference backend
Vision-Pipeline — LLaVA + Moondream image understanding

Image Generation

Stable Diffusion — AUTOMATIC1111 WebUI, text-to-image, inpainting, model switching, GPU-accelerated

Audio & Speech

Whisper (Speaches) — real-time speech-to-text, GPU-accelerated, OpenAI-compatible API

Frontends & Chat

SillyTavern — multi-model chat frontend
Open WebUI — ChatGPT-style web interface

Automation & Orchestration

n8n — workflow automation, 400+ integrations
Flowise — drag-and-drop AI agent builder
Aider — AI pair programmer wired to your local models

Data & Search

Qdrant — vector database for RAG & embeddings
SearXNG — private metasearch engine (no tracking)

Code & DevOps

Tabby-Coder — Qwen2.5 code completion, GPU
VS Code Workspace — dev container, talks to every service
Glances — real-time system telemetry dashboard

Security & Access

Authentik — identity provider, SSO, MFA, RBAC
PostgreSQL + Redis — auth backend infrastructure
Tailscale — zero-config VPN, remote access from anywhere

Extensible by Design

Anything else Docker can run — add it to the compose file and it's live. The stack is designed to grow.

WHAT THIS MEANS FOR YOUR BUSINESS

Replace 12+ subscriptions: one stack covers code completion ($20/user/mo), chat ($20/user/mo), image generation ($10-30/mo), transcription ($0.006/min), workflow automation ($50+/mo), vector DB ($73+/mo), and identity provider ($1-5/user/mo). Breakeven on a 20-person team: under 3 months.
Every modality, one box: text, images, audio, code — your team doesn't leave the network to use AI. No API keys to rotate, no rate limits to hit, no usage-based billing surprises.
Data stays home: proprietary code, internal documents, voice recordings, and generated images never leave your building. Compliance-ready by design.
Latency that feels local: sub-100ms completions and instant image generation beats any cloud round-trip. Your developers stay in flow.
Zero vendor lock-in: model weights, configs, Dockerfiles, and full documentation are yours. You own it forever, or we maintain it for you.
Remote-ready: Tailscale integration means your team can access the full stack from anywhere — securely, without exposing ports to the internet.

WHO DELIVERS IT

Built by Jeremy Panasuk — 22+ years of enterprise data & infrastructure delivered to JPMorgan Chase, PepsiCo, Federal Reserve, Navy Federal, BNY Mellon, Micron, Toyota, and Harvard. $2M+ program management. 19/20 crisis accounts rescued. This isn't a weekend project: it's the same rigor applied to Fortune 100 engagements, packaged for your business.

"You sent one of the best that Informatica has to offer in Jeremy." — MF Global Infrastructure

A COMPLETE PRIVATE AI PLATFORM.