DennyModelsMade in Russia

DENNY · LOCAL AI

Local models.
AI on your device.

Models for Denny on Android and Mac. Compare file sizes, choose a compatible option and try a text conversation without internet.

With a local model selected, text is processed on your device. Cloud AI, web search and external speech services may send data to their respective providers.

ANDROID · LOCAL AI

Gemma 4 E2B

≈ 3.35 GB GGUF · Q4_0

A model for Denny on Android. After downloading it, text can be processed on your phone without connecting to a cloud language model.

Install it through the local AI section of the app. The available catalog depends on your Denny version.

This is the model file size. Running it needs additional memory; speed depends on your phone and conversation length.

MAC · THREE MODELS

Gemma 4 E2B
Bonsai 27B
Gemma 4 12B

≈ 3.35–6.98 GB

Gemma 4 E2B — Q4_0 GGUF, about 3.35 GB.
Bonsai 27B — Q1_0 GGUF, about 3.80 GB.
Gemma 4 12B — Q4_0 GGUF, about 6.98 GB.

Choose in Denny’s local AI settings. Try a compact model first and compare quality and response time on your own tasks. File size is different from memory use.

Check download availability ↗

Catalog entries and file sizes checked on 5 November 2026. Compatibility depends on the app build and its runtime. The current full Denny build requires Apple Silicon M1 or later and macOS 14 or later.

MAC · NOT IN THIS CATALOG

Gemma 4 E4B

This member of the Gemma family is not listed in Denny’s current automatic download catalog.

No Denny download file size or tested memory requirement is listed here for E4B. For a custom GGUF file, check compatibility with your installed runtime.

About the Gemma family at Google ↗

Choose with a small test

  1. Start with the compact model recommended by your installed app.
  2. Try three everyday tasks: a short question, a plan for the day and a draft message.
  3. If the quality is insufficient, try a larger compatible model while leaving memory available for the system.
  4. Compare response time and device temperature. Longer conversations use more memory.

Bigger is not always more convenient

Parameter count is one factor. The model family, quantization, context and runtime also matter. A larger model may handle difficult work better while taking longer to respond.

Your own GGUF file

You can import a compatible file. A GGUF extension alone does not guarantee support: the model architecture and format must work with the app’s runtime. Check the source, license and checksum.

How much memory does local AI need?

A GGUF file’s size is different from RAM use. Memory is also needed for conversation context, the app and your operating system. A 3.35 GB file does not mean a phone needs exactly 3.35 GB of RAM.

Start a model in Denny

  1. Open the local AI section in your installed app.
  2. Select a compatible model in its catalog and check free storage.
  3. Download while connected to the internet.
  4. Switch to local mode and try a short text request with the network off.

Get the current Mac build through AFTEAM support. New Android sales are being prepared; existing owners can get the APK and updates through support. Follow the Denny setup guide for current access and requirements.

Does voice work locally too?

The language model handles text. Speech recognition and speech output are separate components. Their offline availability depends on your settings and installed resources.

Guide: AI without internet →

DENNY · IN PRACTICE

Choosing a model is the beginning

AI without internet

What runs on your device and how to check local mode.

Read ↗

AI on your phone

Android or iPhone, first settings and one useful task.

Read ↗

A personal AI assistant

Tasks, notes and reminders: what to check when choosing.

Read ↗