Did this land?
Off Grid AI Desktop · macOS

The chief of staff
you never had.

Off Grid AI Desktop sees your day, remembers all of it, and gets ahead of you. The meeting you half-remember, the task you promised on Tuesday, the thread you need right now. It has them ready before you ask. Hand it work the way you would a chief of staff, and it takes it from there. It never forgets, and it never logs off, all on your own Mac.

macOS, Apple Silicon · signed and notarized · free to download

Want new features first? Get the nightly build · pre-release, updated on every change, expect rough edges

AI gateway · chat · vision · image · voice · projects · artifacts · connectors · auto meeting recording · dictation · clipboard history · screen memory · a CRM that builds itself · search · actions you approve · and more

You forget what you did last Tuesday. Your Mac does not. Off Grid keeps a quiet record of your work you can actually ask, and it hands you what you need before you go looking. The record runs on your own hardware, so it stays yours.

What you get for free

A full AI studio, on-device. The open core does what a cloud chat app does, without the cloud.

Chat
Text and vision, streaming, with a reasoning mode. 15-30 tokens a second on a flagship Mac.
Image generation
Text-to-image and image-to-image on your own GPU. Ships Z-Image-Turbo and SDXL-Lightning, with a live per-step preview.
Voice
Speech-to-text with Whisper, text-to-speech with Kokoro, and a hands-free voice mode. All on-device.
Projects
Drop in PDFs, docs, and notes, then chat grounded in them with cited sources.
Artifacts
The model's HTML, React, SVG, and Mermaid render live in a sandboxed canvas next to the chat.
Connectors
Add any MCP server and use it inside chat. A preset catalog is built in.
Any model
A curated catalog plus direct Hugging Face search. Qwen, Gemma, Llama, Mistral, or any GGUF.
Offline by default
No account, no API key, no sign-in. Works in airplane mode, on restricted networks, anywhere.

One local endpoint that speaks OpenAI

Point any OpenAI client at http://127.0.0.1:7878/v1 and it works. No key. Chat, vision, image, audio, and embeddings, all on one local server.

from openai import OpenAI
client = OpenAI(base_url="http://127.0.0.1:7878/v1", api_key="not-needed")
print(client.chat.completions.create(
    model="local",
    messages=[{"role": "user", "content": "Hello"}]
).choices[0].message.content)

Run it headless with --server-only for a homelab box, a server, or wiring local models into your own apps.


Off Grid AI Pro: the memory layer

The free app runs models. Pro adds the always-on layer that sees your work, remembers it, reflects it back, and acts with your approval. Opt-in, with a recording indicator. Nothing leaves the device.

It sees
Captures your screen, reads it on-device, and files it into observations and entities. Pixels never leave your disk.
It remembers
Your Day as a journal you can scroll, and Replay, a movie of your day you can scrub frame by frame.
It maps your world
A private CRM that builds itself. The people, projects, and companies across everything you touch, each with a running summary of what is open. You never update it. It updates itself.
It reflects
Where your attention actually went. Mind-share, focus versus context-switching, and how your day and week really broke down.
Meetings
Records and transcribes Google Meet and Zoom locally, then folds the title, summary, and attendees into your timeline.
Dictation
Hold the dictation key, talk, and the text lands at your cursor in any app. Transcribed on-device.
Clipboard
A searchable history of everything you copy, with image and file previews, stored on your disk.
One search
Your screens, meetings, clipboard, and memory in one place. Search your whole work life, readable only by you.
You delegate, it acts
Hand it a task and it understands what you mean. It drafts the reply, files the ticket, lines up the next move. Nothing runs without your approval.
Off Grid AI Pro is live: $49/year. The price climbs as we grow — never down — so today's tier is the lowest it will be. One license covers up to 5 devices, on desktop and mobile.

Why you can trust it

Privacy first by architecture. Your data never leaves your device.

Your local database is encrypted at rest. There is no account to create and no server to leak. 100,000+ downloads across the apps, 2,500+ GitHub stars, a 500-strong community.


Questions

Is it really free? The full studio is free and open source under AGPL. Pro is the paid layer.

Does it work offline? Yes. All inference is local. Airplane mode, restricted networks, anywhere.

Which Macs? macOS on Apple Silicon, M1 and later. Signed and notarized.

Does it phone home? No cloud inference, no account, no API key. Capture is opt-in, with a visible indicator.

What models can I run? Qwen, Gemma, Llama, Mistral, and any GGUF from Hugging Face.

What does Pro cost? $49/year today, climbing toward $99/year as we grow — the tier you join at is the rate you hold. Prefer to own it? $69 once today for lifetime — about a year and a half up front, then yours forever — and it climbs with the ladder too. One license covers 5 devices.