> ## Documentation Index > Fetch the complete documentation index at: https://developers.webai.com/llms.txt > Use this file to discover all available pages before exploring further. # AI > Choose the local model, see Personas on this device, and save facts a later chat can use. AI is the model Chat and Personas use on this computer, the Personas on this device, and the facts saved as long-term memory. Click your name at the bottom of the sidebar, then **Settings**, then **AI**. The page opens on a portrait of the intelligence on this device, then three rooms: **Model**, **Personas**, and **You**. ## Choose a model Picking a model does not download it yet. **Apply** starts the download and load. **1. Open AI** Click your name at the bottom of the sidebar, then **Settings**, then **AI**, then **Model**. **2. Open the catalog** Click the model field. Pick a family, then a size. Some models also ask you to pick a quantization. Models are listed by their catalog display names, not raw repository ids. Each row shows whether that model can accept images. **3. Check that it fits** Each choice shows how much memory it needs. A laptop mark means that model is already on this computer. A download mark means it is not. The footer shows free disk space. **4. Apply** Click **Apply**. Progress appears in the header as **Downloading model** or **Loading model**, with a percent. A model published in several parts shows which part is in progress. If that model is already loaded, **Apply** stays off until you pick a different one. While the same model is downloading, click **Apply** again to stop it. > [!NOTE] > > **Thinking mode** appears when the selected model can show its reasoning. It starts on. Leave it on to get **Thinking** and **Thought process** in [Chat](https://developers.webai.com/using-webai/chat.md). Turn it off to skip that reasoning step. Your [device tier](https://developers.webai.com/settings/system.md#device-tier) sets the model webAI starts on: * **Tier 1** starts on TwIL 3B Pro. * **Tier 2** and **Tier 3** start on Meridian 1 12B. If Meridian 1 12B isn't available, webAI starts on Gemma 4 12B. * **Tier 4** starts on Gemma 4 31B. A Persona that cannot run on the loaded model is unavailable in [Personas](https://developers.webai.com/using-webai/personas.md) until you apply a model that fits. ## Available models The catalog lists these families and sizes. A size your Mac doesn't have the memory for can't be picked. | Family | Sizes | Accepts images | | - | - | - | | Gemma 4 | 31B, 26B-A4B, 12B, E4B, E2B | Yes | | Phi-4 | 14B | No | | Granite 4.2 | 8B | No | | Gemma 3 | 27B, 12B, 4B | Yes | | TwIL | 3B Pro, 3B, 2B, 1.7B | No | | Meridian 1 | 12B | No | Meridian 1 12B is text-only. To work with an image, apply a Gemma 4 or Gemma 3 model. > [!NOTE] > > Meridian 1 appears in the catalog only when it is available on this Mac. ## Idle memory On Macs with **16 GB** of memory or less, the chat model frees its memory after **5 idle minutes**. The next reply takes about **2 seconds** longer to start while the model loads again. The model that searches your documents also frees its memory after **5 idle minutes** on many Macs. If the chat model stops unexpectedly, webAI starts it again on its own. You do not have to restart the app. ## Personas **Personas** on this page lists Personas on this device. Local Personas use the model you applied above. For chatting with a Persona and the full create flow, see [Personas](https://developers.webai.com/using-webai/personas.md). ## You **You** shows **What webAI knows about you**: facts a later chat on this device can use. Some are ones you add. Others are learned in chat. A Persona that runs on someone else's device does not get them. The list holds up to 100 memories. Memory is not used with Gemma 4 E2B, Gemma 3 4B, or TwIL. With one of those models applied, the header shows **Not active for this model**. Your saved memories are kept, and you can still manage them. ### Add a memory **1. Write the fact** Type a short fact in the field at the top of **You**. When the list is empty, click a **Try** starter to begin a sentence. **2. Choose its kind and when it is used** Pick **About you**, **How you like things**, or **Your work**. Click **Always use** to include it in every chat. Leave it off to use the fact only when it is relevant. **3. Save it** Click **Remember**, or press **Enter**. The fact appears under its kind, marked **Set by you**. > [!NOTE] > > webAI is built to decline passwords, keys, payment-card numbers, and health details as memories. Do not rely on that filter for anything sensitive. ### Change or forget a memory The mark beside a memory switches it between every chat and only when relevant. Open a memory's options and click **Edit** to change the text, or **Forget** to remove it. **Undo** brings back a memory you just forgot. **Forget everything** removes every memory on the list after you confirm. It cannot be undone. It does not clear greeting history. Greeting weather is on [General](https://developers.webai.com/settings/general.md). ## Common questions > [accordion] Does selecting a model download it? > > No. Picking a model in **Settings → AI** does not download it yet. Click **Apply** to start the download and load. Progress appears in the header as **Downloading model** or **Loading model**. > [accordion] What is Thinking mode? > > **Thinking mode** appears when the selected model can show its reasoning. It starts on. Leave it on to get **Thinking** and **Thought process** in [Chat](https://developers.webai.com/using-webai/chat.md). Turn it off to skip that reasoning step. > [accordion] What is long-term memory? > > Under **You**, **What webAI knows about you** lists facts a later chat on this device can use. The list holds up to 100 memories. Memory is not used with Gemma 4 E2B, Gemma 3 4B, or TwIL. > [accordion] What does Always use do? > > A memory with **Always use** on is included in every chat on this device. With it off, the memory is used only when it is relevant to the conversation. Click the mark beside a memory to switch it. > [accordion] What does Forget everything do? > > **Forget everything** removes every memory on the list after you confirm. It cannot be undone. It does not clear greeting history. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.