Built-in Private AI
A free AI model that runs on your own computer
Improve a block, grade a prompt or chat, with nothing you write leaving your computer. No API key, no account, no internet connection needed. It's included in PromptMason on every plan.


Private, by default
- No API key and no account. Nothing to sign up for and nothing to pay per use.
- Nothing you write leaves your computer. The model runs on your own computer.
- Works offline. After the one-time download it needs no internet connection.
- Free on every plan, Free and Pro alike.
What you can use it for
AI assist
Rewrite or draft a block, prompt or document while you edit it. You see exactly what changed, and keep it or not.
Evaluate
Grade a prompt out of 10 with notes on what to improve, or check it for personal, financial or confidential details before you paste it into a public AI.
Chat
Run your saved prompts in a conversation and read the replies, with your library a click away.
Qwen3.5 2B is small, so it suits short, focused tasks best. For a deeper read of a prompt, choose one of the larger models below. You can also add your own key for OpenAI, Anthropic, Google Gemini or xAI and pick the model for each run. Keep the built-in model for anything private.
Choose the model that fits your computer
PromptMason offers three models. Qwen3.5 2B is the one you get unless you choose another. Qwen3.5 9B and Gemma 4 12B are larger. Only one is on your computer at a time. To try another, remove the one you have, then download the other.
| What it needs | Qwen3.5 2B | Qwen3.5 9B | Gemma 4 12B |
|---|---|---|---|
| Download | 1.28 GB | 5.68 GB | 6.98 GB |
| Graphics card | Not needed | 8 GB of its own memory | 12 GB of its own memory |
| Memory | 8 GB | 16 GB | 16 GB |
| Free disk space | 3 GB | 12 GB | 14 GB |
In our tests, a larger model gave a deeper read of a prompt than Qwen3.5 2B. It found more of the real problems in a prompt and gave more suggestions you can act on, and it kept every {{variable}} in every test. That is what we saw in our own tests. Your prompts and your computer make a difference.
The two larger models need a graphics card with its own memory. PromptMason can't check your card, so it asks you to confirm before the download starts. Without a suitable card they still run, but much more slowly. How fast any model answers depends on your computer and your graphics card.
Your choice stays on the computer you made it on, so a desktop and a laptop can use different models.
What it needs
- A separate download, started from Settings, then AI & connected apps, only if you want it. Qwen3.5 2B, the model you get unless you choose another, is about 1.3 GB.
- A computer with at least 4 processor cores and 8 GB of memory. Qwen3.5 2B needs no graphics card.
- On a smaller or older computer, rewrites take a little longer.
- With Qwen3.5 2B on a mid-range laptop with 16 GB, the median time to the first words is about half a second and the median whole rewrite takes about 3 seconds; other computers will differ.
The models are Qwen3.5 2B, Qwen3.5 9B and Gemma 4 12B. Qwen3.5 2B and Qwen3.5 9B are made by Qwen, and Gemma 4 12B is made by Google. Each is shared under the Apache License 2.0. They run on llama.cpp (MIT License). See licenses.
For choosing a model, the download step by step, and what to do if it doesn't start, see the local model guide. For what each AI feature sends, and where, see What AI features send.