---
title: "MediSync Local: model packages and offline reserve"
summary: "Which model package Dictate downloads for your computer, where it is stored, and how the offline reserve backs up MediSync Cloud."
url: "https://help.medisynchealth.ai/en/dictate/local-model-packages"
product: "MediSync Dictate"
platforms: ["windows","macos"]
access: "public"
updated: "2026-11-11"
version: "Dictate 0.4.0-beta.6"
---

# MediSync Local: model packages and offline reserve

Which model package Dictate downloads for your computer, where it is stored, and how the offline reserve backs up MediSync Cloud.

MediSync Local needs two models on the computer:

- MediSync Speech for speech recognition, 671 MB.
- MediSync S2 for Smart Dictation. The same package holds the model the assistant uses on this computer to write texts and change selected text.

Dictate picks the variant of MediSync S2 by the graphics. You do not choose anything.

**Windows**

| Own graphics memory | MediSync S2 | Whole package |
| --- | --- | --- |
| 6 GB or more | graphics card variant, 2.1 GB | 2.8 GB |
| 3.5 to under 6 GB | compact variant, 1.4 GB | 2.1 GB |
| less, or no graphics card | not installed, rules format the text | 0.7 GB |

Supported are NVIDIA through CUDA, and AMD and Intel through Vulkan. Only the card's own memory counts. Integrated graphics borrow system memory, which does not count.

**Mac**

A Mac with an Apple chip shares its memory between processor and graphics. What counts is the part macOS gives to the graphics.

| Memory for graphics | MediSync S2 | Whole package |
| --- | --- | --- |
| 10 GB or more, usually Macs with 16 GB of memory or more | graphics card variant, 2.1 GB | 2.8 GB |
| 4.5 to under 10 GB, usually Macs with 8 GB of memory | compact variant, 1.4 GB | 2.1 GB |
| less | not installed, rules format the text | 0.7 GB |

Without MediSync S2, Dictate writes with rules. The assistant is then not available with MediSync Local. In that case the bar shows `Editing model missing`

After loading, Dictate tests MediSync S2 with a short sentence. If the model takes longer than 4 seconds for it, rules format the text. If the model takes longer than 6 seconds for a dictation twice in a row, Dictate pauses it for 10 minutes. During that time rules format the text too.

## Which models are installed

Open **Settings › Status and models**. The **MediSync models** section shows both models with their size and **Installed** or **Not installed**. For MediSync S2 it names the variant: **Formatting, graphics card variant**, **Formatting, compact variant** or **Not supported on this device: rules format the text**. The **This computer** section below compares your computer with the requirements.

## Download the models

The download needs a sign-in with your MediSync account. The models come from `models.medisync.me`.

1. Open **Settings › Status and models**.
2. Under **MediSync models**, click **Download**.

Dictate shows the progress and the time left. You can keep working, the download runs in the background. **Pause** interrupts it, **Resume** continues it. If the connection drops, the download continues where it stopped.

The models stay on this computer only:

- Windows: `%LOCALAPPDATA%\MediSync\Dictate\models`
- Mac: `~/Library/Application Support/MediSync Dictate/models`

Dictate checks every file against a checksum and loads only MediSync's own models. Dictate does not use a damaged or foreign file.

## Offline reserve for MediSync Cloud

When you work with MediSync Cloud, MediSync Local can step in if the cloud is unreachable.

1. Open **Settings › Status and models**.
2. Under **Processing**, at **Offline reserve**, click **Download offline reserve**. Dictate downloads the same models as for MediSync Local.
3. Afterwards, leave **Continue locally when the cloud is down** on.

Dictate checks every 30 seconds whether the cloud is reachable. If it is not, the dictation runs on your computer, and the bar shows `Inserted locally` The reserve does not step in for problems with sign-in or the plan. If your plan's cloud minutes are used up, you keep dictating locally with the reserve until new minutes come.

If you turn off **Continue locally when the cloud is down**, no dictation is possible without the cloud.