The laptop on your desk has enough memory and GPU to run a capable language model, generate images, and transcribe your voice, all on its own. That hardware sits mostly idle while you pay a monthly subscription to do those things on a server you do not control. Off Grid AI Desktop is a free, open-source app that runs the whole ChatGPT-style experience directly on your Mac or PC.
Free, open source, runs offline.
The Off Grid AI Desktop chat, running a local model fully on-device.
What People Use ChatGPT For
Most people use a cloud assistant for a handful of things. They chat and ask questions. They generate images. They talk to it by voice and have it talk back. They upload a document and ask about it. They paste code and get an explanation.
Every one of those sends your words, files, and voice to a remote server, behind an account and a recurring bill. Off Grid AI Desktop does the same set of jobs on your own machine, with no account and no bill.
What Off Grid AI Desktop Can Do
It runs a local language model through a bundled llama.cpp server. You get a chat window with temperature and context-window controls, the same feel as a cloud assistant. Then it covers the rest of what you reach for.
- Chat. Ask questions, draft, summarize, write code. The model runs on-device, so there is no round trip to a server.
- Images. Generate art and mockups with SDXL, SDXL-Lightning for fast results, SD 1.5/2.1, or the 2026 Z-Image-Turbo model. Text-to-image and image-to-image, with a live per-step preview, ETA, cancel, a lightbox, a gallery, and style presets.
- Voice in. A bundled whisper.cpp turns your mic into composer text. Speak your prompt.
- Voice out. Kokoro-82M reads any reply back to you, with an auto-speak mode for a full voice conversation.
- Documents. Upload PDFs, Word files, text, images, or audio. The app builds on-device embeddings and a local vector store, then answers with cited sources. A private chat with your own files.
- Live canvas. When the model writes HTML, SVG, Mermaid, or React, it renders in a sandboxed iframe with no network or file access. Toggle code and preview, then download.
What You Need
The app runs on Apple Silicon Macs and Windows PCs.
| Tier | macOS | Windows |
|---|---|---|
| Minimum | M1, 8GB unified memory, macOS 13, 15GB free | 16GB RAM, GTX 1660 or newer, Windows 10, 15GB free |
| Recommended | M2/M3/M4, 16GB+ unified memory, macOS 14+, 30GB free | 32GB RAM, RTX 3060 12GB or better, Windows 11, 30GB free |
The minimum tier runs chat and small images. The recommended tier handles larger models and SDXL with little waiting.
Which Models to Use
Pick a model that fits your memory. Quantized GGUF files are sized for consumer hardware.
| Your device | Model to try | What to expect |
|---|---|---|
| 8GB Mac / 16GB PC | Gemma 3 4B (Q4_K) | Fast everyday chat and drafting |
| 16GB Mac / 32GB PC | Qwen 2.5 7B (q8_0) | Stronger reasoning and code |
| 32GB+ | A 12B-14B class model | Best quality, slower first token |
Browse and download any of these from the built-in Hugging Face browser, no config file needed.
How Hardware Acceleration Works
On a Mac, the model runs on the GPU through Metal, and Apple Silicon shares one memory pool between CPU and GPU. That unified memory is why a laptop can hold a 7B model and answer fast.
On Windows, acceleration uses CUDA on NVIDIA cards or Vulkan on others, with a CPU fallback that works but runs slower.
Quantization makes it fit. A q8_0 or Q4_K model uses a fraction of the full-precision memory, with little quality loss for chat. That is how chat, images, voice, and document Q&A all run on one normal machine.
Getting a Smooth Experience
Match the model to your memory. If replies crawl, drop to a smaller quantization or model.
Use SDXL-Lightning when you want an image in a few steps instead of full quality. Save the heavier SDXL run for when detail matters.
Trim the context window for short chats. Long context reserves memory you may not use, which leaves less room for speed.
Privacy: Stronger Than a Cloud Chatbot
A cloud chatbot sends your prompts, uploads, and voice to a remote server under a policy you cannot inspect, behind a login.
Off Grid AI Desktop sends nothing. No account, no subscription, no telemetry, no API key. Your chats, images, voice, and documents stay on your disk. The code is AGPL-3.0, open for anyone to read. Disconnect from the network and it still works.
Getting Started
- Open the repo at github.com/off-grid-ai/desktop.
- Download the build for your OS, or clone and build from source.
- Launch the app.
- Open the models browser, pick a model that fits your memory, and download it.
- Chat, generate an image, speak a prompt, or drop in a document.
You have a private, offline ChatGPT-style studio in a few minutes.
What's Coming
- Cross-device sync so your chats and memory follow you between machines.
- More language and image models as they ship.
- Unified search across chats, documents, and captured work.
- A richer day view built from what the app remembers.
FAQ
Q: Is it really free?
Yes. The core app is free and open source under AGPL-3.0. No subscription and no paywall on the features here.
Q: Does it work offline?
Yes. Download a model once, then disconnect. Chat, images, voice, and document Q&A all run on-device.
Q: Do I need an account?
No. There is no sign-up and no login. You download the app and run it.
Q: Which models can I run?
Open-weight GGUF models like Gemma and Qwen for chat, and SDXL, SD 1.5/2.1, and Z-Image-Turbo for images.
Q: How much RAM do I need?
8GB on a Mac or 16GB on a PC runs a small model. 16GB or more is better for larger models and images.
Q: Is my data private?
Yes. No account, no telemetry, no servers. Everything stays on your machine.
A private, offline alternative to ChatGPT: chat, images, voice, and documents, all on your own desktop.
Top comments (0)