Idaus 0.1.0 · Apple Silicon preview

Meet Idaus.

A model on your machine.

Trained on Google's Gemma. Fine-tuned to match the frontier on reasoning and code. Runs entirely on your Mac via MLX — no account, no cloud, no token meter.

macOS 13+ · Apple Silicon 172 MB · .dmg Free during preview
A first.

For thirty years, the smartest AI lived in a building you'd never see. Today, the chain breaks.

You rented access. You paid by the token. You shipped your prompts to a server farm three time-zones away, then waited in line for an answer that was never quite yours. That was the deal. It is no longer the deal.

Idaus is the first model in the Gods Eye family that runs entirely on the silicon already in your hand. Same transformer architecture as the frontier. Distilled. Quantised. Hand-tuned for code and reasoning. Then folded into the chip you bought to check email.

1978 · then

GPS was a Cold-War Navy weapon.

Twenty-four satellites. Three control stations. Designed to land cruise missiles within ten metres of a Soviet bunker.

today · now

The blue dot on your map.

Free. Instant. Open in your pocket while you wait for the bus. Nobody calls it revolutionary anymore. It just is.

1999 · then

LiDAR mapped Martian craters.

Bolted to the side of a NASA rover. Cost the US treasury six hundred million dollars. Every photon was a miracle.

today · now

The autofocus on your iPhone.

Same physics. Apple shrunk it to fit next to the camera lens. Now it focuses your kid's birthday cake in a dark room.

2017 · then

The transformer was a Google paper.

Eight pages, "Attention is all you need." Required a thousand-GPU cluster to train. Lived behind an API and a credit card.

today · now

Idaus, on the chip under your palm rest.

Same architecture. Same capability class for everyday work. Runs offline. Owes you nothing. Costs nothing.

This is what it sounds like when something stops being expensive,
stops being far away, and starts being yours.

PROMISE 01

Truly private.

Your prompts, your context, your code never leave the machine. No telemetry. No cloud round-trip. Idaus runs in the silicon, full stop.

PROMISE 02

Surprisingly fast.

Native MLX inference on Apple Silicon. First token in ~150ms on M2, 30+ tokens/sec sustained. No spinner, no rate limit, no waiting in line.

PROMISE 03

Zero token meter.

Free during preview. Run it on a flight, behind a firewall, all night, every night. No API key, no billing, no tier you'll forget you're on.

9B
parameters
8K
context
5.5GB
on disk
8GB+
RAM needed

Install in three steps.

macOS · Apple Silicon
01

Download the .dmg

Click the download button at the top. The file is Idaus-0.1.0-arm64.dmg — about 172 MB.

02

Drop in Applications

Double-click the .dmg, drag Idaus into your Applications folder. Eject the disk image. Standard macOS rhythm.

03

Approve once in Terminal

Idaus is ad-hoc signed (no Apple Developer ID yet). Open Terminal, paste the line below, hit . The app launches normally from then on.

$xattr -dr com.apple.quarantine "/Applications/Idaus.app"
What's coming

This is 0.1.0.
The journey just started.

Q3 · this year

Windows + Linux

Idaus on llama.cpp + Vulkan. Same model, same quality, every laptop you own.

Q4 · this year

Idaus Pro

27B mixture-of-experts. Frontier-tier coding and analysis. Still private, still local.

Always

Open weights

Inspectable, fine-tuneable, distillable. The model is the artifact, not the secret.

Quick answers.

Why is the app called Idaus?
Idaus is a working name for our home-grown model line. It draws from Latin / Old-English roots — the "watchful" and "true" senses — and pairs deliberately with the Gods Eye family. Expect the name to stay.
Is it really comparable to Sonnet 4.6 / GPT-5?
On internal coding and reasoning evals at this 9B size, yes — for everyday tasks. For the heaviest multi-hour reasoning runs, the frontier cloud models still pull ahead. Idaus's strength is the 90% case where you want quality, privacy, and zero latency.
Why does macOS warn me on first launch?
We're shipping ad-hoc signed during preview while we set up our Apple Developer ID + notarisation pipeline. The Terminal command above strips the quarantine flag — it's the same workaround you'd use for any small indie Mac app. Once we're notarised, this step disappears.
Will my data be used to train future versions?
No. Idaus runs entirely on-device. We don't see your prompts, your code, or your context. Future training data will come from explicit, opt-in research partnerships — never from the app.
How do I use Idaus inside Gods Eye?
After the install, restart Gods Eye. Idaus appears in the agent picker as a local backend. Pick it from the dropdown and chat — the desktop app handles the MLX runtime, model load, and streaming.