Free, private AI on your PC: the local model

PromptMason includes an AI model that runs on your own PC, free on every plan, with no API key or account. Download it, make it your default, and fix download or start-up problems.

PromptMason includes its own AI model, the local model, that runs entirely on your computer. Once you’ve downloaded it:

• It’s free. It’s included on every plan, including Free. There’s no API key to get, no account to create and nothing to pay per use.
• It’s private. Your text is processed on your PC. None of it is sent anywhere.
• It works offline. After the one-time download, it needs no internet connection.

What you can use it for

• AI assist: rewrite or draft a block, a static prompt or a document while you edit it. See Improve text with AI assist.

  

• Evaluate: grade a block, prompt or document, or check it for personal, financial or confidential details before you share it anywhere. See Evaluate a block, prompt or document.

• Chat: run your prompts and read the replies on the Chat & Browse screen, and let the chat find and use things in your library. See Run prompts in Chat.

The local model is a small model, so it suits short, focused tasks best. For long text or harder tasks, a cloud AI provider’s larger models can do more. You can use both: pick the model for each run, and keep the local model for anything private. See Set up AI and Choosing what to share with AI.

What it is

• The model is Qwen3.5 2B, shared under the Apache License 2.0, and runs with llama.cpp (MIT License). Once it’s downloaded, Settings → General → About PromptMason lists both.
• The download is about 1.3 GB. The card shows the size before you download, and the space it uses once it’s here.
• It’s a small model. If text is too long for it, a message says so, and a cloud provider’s larger models can handle it instead (see Set up AI).
• In model lists it’s called Local model and has a lock icon. Pointing to it shows Runs on this computer. Nothing is sent anywhere.

What your PC needs

• At least 4 processor cores and 8 GB of memory. On a smaller PC, the card says, for example, This computer has 2 processor cores and 4 GB of memory. The local model needs at least 4 cores and 8 GB. and there’s no Download button. Use an AI provider instead. If the local model is already your default, the card also says Choose another provider under Default AI provider below.

  

• A PC with less than 16 GB of memory, or fewer than 4 physical processor cores, shows Rewrites may take a little longer on this computer. on the card before you download. The model still works there.

• Free disk space for the download. If there isn’t enough, the download stops with a message that says how much space it needs.

Download it

1. Click Settings in the sidebar, then AI & connected apps. (On the Chat screen, the Download the free local model button opens this pane at the card.)

2. The first card is Local model (private). It shows Not downloaded. Click Download.

3. The Download the local model window says which license the model is shared under. Click Read the Apache License 2.0 to read it, then click Accept and download. This window only appears the first time.

   

4. The card shows Downloading, a progress bar, the amount downloaded so far, the speed and the time left.

5. At the end it shows Checking and Checking the download… while PromptMason makes sure the file is complete and undamaged.

When it’s done, the card shows Ready and It uses about 1.3 GB on this computer.

You can leave Settings and keep working while it downloads. If the connection drops or the PC sleeps, it retries a few times on its own and carries on from where it stopped. If you quit PromptMason, or it stops trying, the part already downloaded is kept: open the card and click Resume (or Try again) to carry on.

Make it the default

On the card, click Use as default. AI assist, Evaluate and new chats then start with the local model. This also turns on the AI assist switch in Settings → Features if it was off.

• The card then says AI assist uses the local model.

  

• If it says AI assist is turned off., click Turn on AI assist.

You don’t have to make it the default to use it. You can choose Local model from the Model list each time you run AI assist, Evaluate or a chat. When you choose Privacy in Evaluate, the local model is picked for you if it’s downloaded.

Pause or cancel the download

1. While it downloads, click Cancel.
2. The card shows Paused and how much is downloaded, for example Downloaded 412 MB of 1.3 GB. The part already downloaded is kept.
3. Click Resume to carry on, or Remove to delete the part downloaded. Remove here doesn’t ask first.

What the card shows

The model starts the first time you use it (or when PromptMason opens, if it’s your default). It stops after a while with nothing to do, and when you quit PromptMason, so it only takes up memory while you use it. The first answer after it starts takes a little longer.

After you use AI assist with it, the card shows how long the last rewrite took (a line starting Last rewrite:). Evaluate and Chat don’t update it, and the line goes away when you close PromptMason.

Remove it

1. On the card, click Remove.
2. The Remove the local model? window says: This deletes the downloaded model from this computer. You can download it again later. If the local model is your default, it also says which provider AI assist will use instead, or that AI assist won’t work until you download the model again or add an API key.
3. Click Remove.

Updating PromptMason keeps the local model. Uninstalling PromptMason deletes it, so after a reinstall you download it again. See Install, update and remove.

Messages and what to do

Download and remove messages

These show on the card when the download can’t start or stops, or the model can’t be removed.

Start-up and use messages

These show on the card, or where you used the model (in the AI assist error message, or in a chat reply).

When a start-up problem shows on the card, Details may appear under it. Click Details, then Copy details, and paste the details into an email to support@promptmason.ai if you need help.