Did this land?

Off Grid AI - Private AI. No cloud. No compromise.

Off Grid AI

A private AI that runs on the hardware you already own.

Chat, images, vision, voice, documents - on iOS, Android, macOS, and Windows, running on hardware you own. Local models keep your prompts on your device. If you connect a remote model, Off Grid AI sends requests only to the server you choose. Off Grid AI Pro adds one intelligence layer across your phone and laptop. The layer that merges the two is landing through July.

Don't take the privacy claim on faith - read the code that keeps your data on your device:

iOS 0.0.103 is live on the App Store. macOS has a stable build. Windows stable 0.0.42 is available now. Windows release notes.

100k+
downloads
Play Store, App Store & GitHub
2,900+
stars on GitHub
500+
strong community

Over 100,000 people already run AI on their own phone with Off Grid AI. No account, no subscription for the core, no cloud. The phone in your pocket has enough compute to run a capable model offline, at real speed - Off Grid AI makes it do exactly that.


Start with the app

Capability Details
Text generation Llama, Qwen 3, Gemma 3, Phi-4, Mistral and any GGUF model - 15–30 tok/s on flagship devices
Image generation On-device Stable Diffusion - 5–10s on NPU (Snapdragon), Core ML on iOS. 20+ models
Vision AI Point your camera at anything and ask questions. SmolVLM, Qwen3-VL, Gemma 3n
Voice input On-device Whisper speech-to-text. Hold to record, auto-transcribe. No audio leaves your phone
Tool calling Web search, calculator, date/time, device info. Automatic tool loop
Document analysis Attach PDFs, CSVs, code files. Native PDF text extraction on both platforms
Remote servers Connect to Ollama, LM Studio, LocalAI on your home network
Works offline Airplane mode, restricted networks, anywhere

Where this is going

The app is the first piece. The whole is a Personal AI OS: a private intelligence layer that lives across your phone and your laptop, learns your day in the background, and gets ahead of you the way a chief of staff would.

Your phone knows your life. Your laptop knows your work. Today neither has the full picture. Off Grid AI unifies them into one working model of who you are and what you are doing. It syncs over your own network, never a cloud relay. It does not wait to be opened - it briefs you on the day, surfaces the item you left open, and drafts the reply before you remember you owe it.

Nothing is sent anywhere, because there is no server to send it to. It is open source, so you can check.


Off Grid AI Pro: it was in the room

The free app runs models. Off Grid AI Pro adds the layer that sees your day, remembers it, and gets ahead of you, the way a chief of staff would. Always on, on your own hardware. It is live now on your laptop and your phone. You never brief it. It briefs you.

Capture, on device
It quietly takes in what you see and do, on your screen and in your meetings, and turns it into memory. Opt-in per device, with a recording indicator. The pixels never leave your disk.
One search across everything
Find the message, the page, the meeting, the screen from last week, all in one place that only you can read. The end of where did that go.
A CRM that builds itself
It maps the people, projects, and companies you work with, and what is open with each. A relationship graph you never have to update.
Replay your whole day
Scrub through your day like a recording. Every screen, in order, on device. The thing you swear you saw three hours ago is right there.
Where your time really went
You start no timer and tag nothing. At day's end the breakdown is just there: 4h on the rewrite, 50m on Slack, 30m you will wish you had back.
It acts, you approve
It drafts the reply, files the ticket, finds the commitment you made in an email. Every action is a proposal you approve, and every approval is logged.

Why local AI matters

When you run a query on a cloud AI service - ChatGPT, Gemini, Claude - it’s logged on a server. Your prompt, the response, the time, your account. Stored indefinitely. Used to train future models. Subject to law enforcement requests. Readable by employees.

With Off Grid AI, none of that applies. The model runs in your phone’s memory. Inference happens on your CPU and GPU. Nothing is sent anywhere. Ever.

Privacy here isn’t a setting or a promise. It’s the default output of the architecture. The system has no mechanism to do otherwise - and because the code is open, anyone can verify it.


Get started

Guides

LLMs

Image Generation

Vision, Voice and Documents

Tools and Intelligence

Remote Servers


Community

Questions, feedback, and feature requests - join the Slack community.

Source code is open - star the repo on GitHub.