Nido Español Investors

Coming soon to iPhone

The AI that lives on your iPhone.

Nido is a private AI assistant that runs a 35-billion-parameter model on the iPhone itself. No cloud, no account, no connection needed.

Ask it anything.

Assistants that run inside a phone usually use models of around 3 billion parameters. Nido’s has 35 billion.

The first word in 1.6 seconds.

And the rest at 7.6 tokens per second: the speed measured on an iPhone 16 Pro Max, the same one you see on screen.

You always know what it’s doing.

The egg fills up while Nido reads your message, then follows each word as it writes.

A screen from Nido’s design demos; the answer is a sample.

Privacy

What you tell Nido stays on your iPhone. There’s no account to create and no server to receive it, so nothing can leak and nothing can be used to train anyone’s AI.

  • No account

    No email, no password. Open the app and start typing.

  • No cloud

    Your conversations are stored only on your iPhone.

  • Offline

    It answers just the same on a plane, on the subway or wherever there’s no signal.

Features

A complete assistant that never leaves the phone.

Chat already works in the app. The rest is designed and in development: these screens are its demos.

In the app today

Write, ask, let it reason.

Chat without limits and, with Nido Pro, let it reason before it answers. From the “Add” glass you’ll be able to attach photos, documents and files from your Mac.

Coming soon

Talk to Nido.

You speak and Nido answers out loud, full screen. Listening and speaking happen on the iPhone too: the audio is never stored and never leaves the phone.

Coming soon

Ask it to set things up.

Nido prepares reminders, events and notes, and you confirm each one on a card. It never does anything without your permission.

Coming soon

With Siri and the Action button.

Ask Siri to have Nido set up a reminder without opening the app, or start talking to Nido from the Action button or Control Center.

Coming soon

It remembers what you ask.

“Remember that I’m vegetarian.” Nido stores it only on this iPhone, and you can change or delete it whenever you like.

Ovum, Nido’s engine

25 GB of model.
8 GB of memory.

Nido’s model takes up about 25 GB, and the iPhone 16 Pro Max has 8 GB of memory. Ovum, Nido’s inference engine, keeps only what’s essential in memory and reads the rest from storage as it writes.

Qwen3.6-35B-A3B · 40 layers × 256 experts Already in memoryRead from storage
Each row is one of the model’s 40 layers, and each dot one of its 256 experts. To write a token, Nido goes through 8 experts in every layer. 92–93% of them are already in memory when they’re needed, because Ovum learns which ones you use most; the rest are read from storage at 1.74 GB/s. Shown here in slow motion: on the iPhone, a token takes 132 ms.

9.3tokens/s

when it starts answering, and 7.6 on average

1.6s

to the first word of a new chat, down from 6.4 s

34s

to read a 3,000-token text, down from 5.7 minutes

Measured on an iPhone 16 Pro Max, with the app opened from its icon. A token is the chunk of text the model works with: about 3.6 characters of Spanish.

Fourteen in-house techniques.

Ovum starts from colibri, an open-source engine that runs huge mixture-of-experts models on computers. To bring it to a phone, all of the GPU code and about 40% of the model code are new.

  • Learning cache
  • Mixed 4- and 8-bit trunk
  • qs16 scales
  • Cache-aware routing
  • Layer lookahead
  • Expert predictorin research
  • Custom Metal executor
  • Layer-by-layer GPU reads
  • Warm start
  • Sampling without sorting the vocabulary
  • WFE-based waiting
  • Thermal governorin testing
  • Lightweight UI while answering
  • An interface that tells the truth

Pricing

Free. And Pro, with a single payment.

No subscription: the model runs on your iPhone, so there are no servers to pay for every month.

Nido

Free

  • Unlimited chat with the model
  • Everything on the iPhone, no account
  • Five actions every 24 hours

PRO

Nido Pro

€25.99one-time purchase

  • Reasoning before answering
  • Document attachments
  • “Summarize and continue” when the chat fills up
  • Unlimited actions

In-app purchase, works with Family Sharing.

Frequently asked questions

What people ask most.

Does Nido work without an internet connection?

Yes. The AI model lives on the iPhone, so Nido answers just the same in airplane mode, on the subway or wherever there’s no signal.

Do my conversations leave my iPhone?

No. They’re stored only on your iPhone, and Nido needs no account and no server to answer. Nobody else reads them or uses them to train models.

Which AI model does Nido use?

Qwen3.6-35B-A3B, a 35-billion-parameter mixture-of-experts model from Alibaba Cloud’s Qwen team. It runs on Ovum, Nido’s inference engine.

How does a 25 GB model fit on an iPhone with 8 GB of memory?

Ovum keeps the model’s most-used experts in memory and reads the rest from storage as it writes. When they’re needed, 92–93% of them are already in memory.

Is it fast?

On an iPhone 16 Pro Max, the first word of a new chat arrives in 1.6 seconds, and Nido writes at 7.6 tokens per second on average (9.3 when it starts answering).

How much does it cost?

Nido is free, with unlimited chat. Nido Pro costs €25.99 as a one-time purchase, with no subscription, and works with Family Sharing.

When is it coming out?

Nido is coming soon to the App Store. Chat already works in the app; the other features are designed and in development.

For investors

From 3.4 to 9.3 tokens per second in three days.

Every Ovum improvement is measured on the same iPhone, with the app opened from its icon and the phone cool. The goal with this model is 10 tokens per second. With Ling-3.0-tiny, a lighter mixture-of-experts model being ported to Ovum, it’s 15–20.

Goal: 10
3.2–3.6Until Oct 9
CPU only
4.2Oct 10
On the GPU
5.5Oct 10
Layer lookahead
6.9Oct 10
Lightweight UI
7.6Oct 11
qs16 and WFE
9.3Oct 11
Bigger cache
Tokens per second in the app, on an iPhone 16 Pro Max. The last bar is the speed over the first 600 tokens of an answer; that test’s average is 7.6.

€0in server costs per answer

The iPhone does the computing.

A cloud AI service pays for the servers behind every answer it gives. With Nido, the user’s own iPhone computes it. That’s why Nido Pro can be sold for a one-time price, and the product keeps working even if cloud prices change.

What’s next.

  1. Done

    Nido answers on the iPhone

    With Qwen3.6-35B-A3B, since October 4, 2026.

  2. Done

    Ovum on the iPhone’s GPU

    Learning cache, layer lookahead and warm start, October 9 to 12.

  3. In progress

    Faster and lighter

    A sustained 10 tokens/s with Qwen, and the move to Ling-3.0-tiny, aiming for 15–20.

  4. Next

    App Store launch

    With the features from the demos, and Nido Pro as a one-time purchase.

Alberto Catalán Frutos

Founder. Designs and builds Nido and Ovum.

Everything stays in the nest.

Nido is coming soon to iPhone.