VoxAI Orchestrator

 ##     ##  #######  ##     ##    ###    ####
 ##     ## ##     ##  ##   ##    ## ##    ##
 ##     ## ##     ##   ## ##    ##   ##   ##
 ##     ## ##     ##    ###    ##     ##  ##
  ##   ##  ##     ##   ## ##   #########  ##
   ## ##   ##     ##  ##   ##  ##     ##  ##
    ###     #######  ##     ## ##     ## ####

A unified local AI workspace with agentic capabilities

Local LLM Chat • Cloud Providers • Image & Video Generation • Agentic Web Search • Remote Access

Beta software: active development, frequent iteration, and occasional breaking changes are normal.

---

What Makes VoxAI Different?

VoxAI is not just a model launcher. It is a local-first AI workspace that combines:

• local GGUF chat
• cloud/provider chat
• local diffusion image generation
• provider image and video generation
• agentic search behavior
• a desktop-managed public WebUI

The goal is to keep a powerful desktop app while also exposing a polished browser front end when you choose to run IronGate.

Open-Source Build Notes

This repository is the public OSS project.

• The desktop app, WebUI, local model support, provider integrations, and remote access flow are included here.
• Private monetization/runtime pieces are not meant to live in the public repo.
• Mutable user/account/payment/runtime state should stay in ignored runtime files only.
• Use example/template configs and keep your real keys in local private files.

In practical terms, this repo should contain the product and tooling, but not your private billing or cloud-storage business runtime.

Core Features

Chat

• Local GGUF chat with llama.cpp/Vulkan-based workflows
• Ollama support
• Provider chat for Gemini, OpenAI, OpenRouter, DeepSeek, Kimi, Mistral, xAI, and Z.ai
• Streaming responses
• Code-aware responses and side-panel extraction
• Multi-turn conversation context
• Agentic search support through the desktop service

Image And Video

• Native local image generation without ComfyUI dependency
• Local support for SD 1.5, SDXL, Pony, and Flux
• Provider image/video generation adapters
• Prompt translation / prompt enhancement helpers
• WebUI image queue, gallery, and mobile-friendly generation flow

IronGate WebUI

• Obsidian-based WebUI layout
• Mobile-specific front-end flow
• OAuth-capable login flow
• Cloudflare-first public tunnel flow with Ngrok fallback
• Desktop-controlled public gateway lifecycle
• Account/profile surfaces for WebUI users

Quick Start

Prerequisites

• Windows is the primary target
• Python 3.10+
• AMD or NVIDIA GPU recommended for local generation
• 8GB+ VRAM recommended for the broader local feature set

Fast Start

git clone https://github.com/AlexC1991/AI_GUI.git
cd AI_GUI
start.bat

start.bat handles:

• virtual environment setup
• dependency installation
• runtime directory bootstrapping
• launcher startup

Manual Setup

python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
copy config.example.json config.json
python main.py

Local Runtime DLLs

For Vulkan-accelerated local inference, place the required llama.cpp runtime DLLs in the project root:

• llama.dll
• ggml.dll
• ggml-vulkan.dll
• ggml-cpu-haswell.dll

Model Setup

Chat Models

Place local .gguf chat models in models/llm/.

Recommended examples:

• Qwen3-8B-Q5_K_M.gguf
• Llama-3.2-3B-Instruct.gguf
• DeepSeek-R1-Distill-8B.gguf

Local Image Models

Typical local image model layout:

AI_GUI/
└── models/
    ├── checkpoints/
    ├── loras/
    ├── vae/
    └── text_encoders/

Supported local image model file types include:

• .safetensors
• .ckpt
• .gguf

Cloud Providers

Provider keys can be configured through the desktop Settings UI or by editing your local config files.

Current chat/provider stack in the project includes:

• Gemini
• OpenAI
• OpenRouter
• DeepSeek
• Kimi
• Mistral
• xAI
• Z.ai
• Ollama

The project also includes media-provider routing for image/video-capable services where configured.

Flux On 8GB VRAM

Current public README state:

• Flux Q2 dev / GGUF path: supported for 8GB-class usage
• Flux FP8 safetensor path: still partial / half-working

Recommended 8GB-oriented files:

1. flux1-schnell-Q4_K_S.gguf in models/checkpoints/
2. t5-v1_1-xxl-encoder-Q4_K_M.gguf in models/text_encoders/
3. clip_l.safetensors in models/text_encoders/

IronGate Remote Access

IronGate can expose your local app through a public tunnel.

Preferred Setup

1. Use Cloudflare as the primary public tunnel
2. Keep Ngrok as the secondary/fallback option
3. Launch the public gateway from the desktop app
4. Keep desktop control enabled so the public server only runs while the desktop app is open

Host Config

Copy:

gateway/host_config.example.json -> gateway/host_config.json

Then configure your tunnel and OAuth settings locally.

OAuth Callback Examples

• https://your-domain.example/auth/google/callback
• https://your-domain.example/auth/github/callback

WebUI Highlights

• provider-backed chat, image, and video surfaces
• mobile-focused WebUI layout improvements
• account/profile panels
• local history and user preference persistence
• theme/accent support

Repo Hygiene

The repo is intentionally set up to keep local/runtime artifacts out of source control.

Ignored or local-only areas include things like:

• model weights and offload blobs
• runtime gateway data
• secret/config key files
• generated outputs
• future memory-manager work under VOXAI_Memory_Manager_Safetensor/

Placeholder keep-files preserve directory structure for:

• outputs/images/
• engine/data/conversations/
• engine/data/vectordb/
• gateway/exports/
• gateway/uploads/
• gateway/uploads_desktop/

Project Structure

AI_GUI/
├── backend/              # Chat/image workers and pipeline orchestration
├── engine/               # Local chat engine and memory systems
├── gateway/              # IronGate FastAPI host, templates, static assets
├── ide_core/             # IDE-side helpers and tooling
├── providers/            # Chat, image, and video provider adapters
├── widgets/              # Desktop PySide6 UI
├── config.example.json
├── main.py
├── main_window.py
└── start.bat

Configuration

Main App Config

Copy:

config.example.json -> config.json

Then add the local/provider settings you want to use.

Web Gateway Host Config

Copy:

gateway/host_config.example.json -> gateway/host_config.json

That host config is where local tunnel/OAuth settings belong.

Dependencies

Main dependency groups in the project include:

• PySide6 for the desktop UI
• diffusers, transformers, and torch for image pipelines
• provider SDK/client layers such as OpenAI and Gemini-related packages
• fastapi and uvicorn for IronGate
• utility packages like psutil, colorama, and related helpers

Install everything with:

pip install -r requirements.txt

Contributing

Contributions, issue reports, and cleanup suggestions are welcome.

License

MIT. See LICENSE.

---

Built for the local AI community

Giving small models big capabilities