Wynner AI
A private AI assistant that stays on your computer.
Wynner AI is a desktop assistant that runs the Gemma 3 12B language model locally. Chat works without an internet connection, and no conversation is sent to a server.
Overview
The problem
Most AI assistants send every conversation to a cloud server. For personal questions, that is not always what people want.
The solution we are building
A desktop assistant that is private by design: the language model runs on your own machine, so conversations stay there.
Who it’s for
People who want a personal AI assistant for everyday questions and prefer their conversations to stay on their own computer.
How it works
Using Wynner AI, step by step.
Step 1: Open Wynner
The app starts its local engine, which loads the language model into memory.
Step 2: Ask a question
Type a message or pick a suggestion card. A “Wynner is thinking…” indicator shows while the model works.
Step 3: Get an answer offline
The answer is generated on your PC. In testing, the first reply after launch took 75–135 seconds while the model loaded.
Step 4: Start fresh
New Chat clears the conversation.
Product showcase
See the current build.
Home screen
A real question, while the local model works
The model’s answer, generated offline
Key features
What it does today.
- 01
Local model
A C++ engine built on llama.cpp loads Gemma 3 12B Instruct (Q4_K_M, 7.3 GB) with a 4,096-token context and 22 layers offloaded to the GPU.
- 02
Desktop interface
A Qt Quick (QML) interface with a chat view, a “Wynner is thinking…” indicator, suggestion cards and New Chat, in a frameless navy-and-gold window.
- 03
Clean replies
A C++ bridge starts the engine, sends each message and reads the answer, removing the engine’s placeholder text so only the final reply appears.
Development progress
What works today, and what doesn’t yet.
Implemented3
Built and visible in the current build
- Local chat with Gemma 3 12B through llama.cpp
- Chat view, thinking indicator, suggestion cards and New Chat
- Frameless navy-and-gold window with custom window controls
Offline-tested2
Checked by automated or recorded tests, without live services
- Real conversations tested offline on Windows on 8 November 2026
- The first reply after launch took 75–135 seconds while the model loaded
Planned3
Not built yet
- Sidebar pages: AI Assistant, AI Models, Documents, Data Insights, Tools and Settings
- Attachments and code tools in the chat box
- A packaged installer
Get updates
Follow Wynner AI as it develops.
Early access registration is coming soon. Until then, progress is posted on the Updates page, and you can message Wynn Orbit directly about Wynner AI.
Latest Wynner AI updates
- Wynner AITested offline
Wynner AI: new design and offline conversation test
Wynner moved to a navy-and-gold design, and replies now show only the model’s final answer. Real conversations were tested offline on Windows; the first reply after launch took 75–135 seconds while the model loaded.