TypingMind×
Llama4How to run Llama4 locally as private, self-hosted AI
Run Llama4 from Meta as your own private, self-hosted AI. A free local AI tool runs the model on your computer, and TypingMind gives you a full chat app on top of it. Your prompts stay on your machine, and there is no per-message cost.
About Llama4: Meta's latest collection of multimodal models.
Llama4 Overview
| Model | Llama4 |
| Made by | Meta |
| Runs on |  Ollama |
| Sizes | 16x17b, 128x17b |
| Download size | 67 GB to 245 GB |
| Context window | 10M tokens |
| Input | Text, Image |
| Capabilities | VisionTool calling |
| Ollama command | ollama pull llama4:16x17b |
| Downloads on Ollama | 1.9M |
| Last updated | Jun 16, 2025 |
| Model icon | Get the icon for TypingMind |
Which size of Llama4 should you pick?
Each size is a different version of Llama4. Bigger sizes give better answers but need more memory. Pick the biggest one that fits your computer's RAM. On a Mac, memory is shared with the graphics chip, so leave some room for other apps.
| Model ID | Download | Suggested RAM | Context | Input |
|---|---|---|---|---|
llama4:16x17bdefault | 67GB | 96 GB | 10M | Text, Image |
llama4:128x17b | 245GB | 512 GB | 1M | Text, Image |
Suggested RAM is a rough guide: the download size plus about 2 GB, with room for your system. Longer context windows use more.
Run Llama4 with Ollama
Ollama is a free local AI tool and the easiest way to run Llama4 on your computer. The steps below use llama4:16x17b, the size Ollama downloads by default. Swap in another model ID from the table if you want a smaller or bigger size.
Install Ollama
Ollama is a free, open-source app that runs AI models on your own computer. It is the engine of your self-hosted AI: it downloads Llama4, loads it into memory, and answers requests on your machine.
- Download Ollama for macOS, Windows, or Linux from ollama.com/download and open it.
- On Linux, you can install it from the terminal instead:
curl -fsSL https://ollama.com/install.sh | sh
Ollama asks you to create an account. You don't need one to run models on your own computer. Click No thanks, I'll use Ollama locally. Ollama keeps running in the background after that, so you can close its window.
Download Llama4 with Ollama
Open the Terminal app (Command Prompt or PowerShell on Windows) and run:
ollama pull llama4:16x17bThis downloads llama4:16x17b (67GB) to your computer. When you see success, the model is ready.
To see what you have downloaded, run ollama list. To see a model's context length and features, run ollama show llama4:16x17b.
Let TypingMind connect to Ollama
By default, Ollama only answers apps running on your computer. To let the TypingMind web app talk to your local AI, add https://www.typingmind.com to Ollama's allowed origins with the OLLAMA_ORIGINS setting. Your messages still go straight from your browser to Ollama on your computer.
macOS
launchctl setenv OLLAMA_ORIGINS "https://www.typingmind.com"Then quit Ollama from the menu bar and open it again. This setting is reset when you restart your Mac, so run the command again after a restart.
Windows
- Quit Ollama from the taskbar.
- Search Windows for environment variables and open Edit environment variables for your account.
- Add a new variable named
OLLAMA_ORIGINSwith the valuehttps://www.typingmind.com, then click OK. - Start Ollama again from the Start menu.
Linux
Open the Ollama service settings:
sudo systemctl edit ollama.serviceAdd these lines, save, then run sudo systemctl daemon-reload && sudo systemctl restart ollama:
[Service]
Environment="OLLAMA_ORIGINS=https://www.typingmind.com"💡 Give the model more memory for long chats (optional)
Ollama often runs models with a short context window (4,096 tokens on our 24 GB Mac), so long chats forget earlier messages. Raise it to 32,768 tokens in the Ollama app's Settings → Context length, or on macOS with launchctl setenv OLLAMA_CONTEXT_LENGTH 32768, then restart Ollama. Bigger context windows use more memory.
Add Llama4 to TypingMind
Open www.typingmind.com in your browser. Use the web version: the TypingMind Mac app and the Setapp version block plain http:// addresses like Ollama's.
Click Models in the left sidebar, then Add custom model.
Fill in the form:
| Name | Llama4 16X17B (local)or any name you like |
| Model ID | llama4:16x17b |
| Context Length | 32768match Ollama's context length |
| API Type | OpenAI Chat Completions API |
| Endpoint URL | http://localhost:11434/v1/chat/completions |
| Authentication Type | No authentication |
Llama4 16X17B (local)llama4:16x17b32768http://localhost:11434/v1/chat/completionsOpen the Capabilities tab and turn off Auto check capabilities on save. The check sends several test requests, and a local model answers them one at a time, so it can take minutes.
Then set the features for Llama4:
- Support System Role: on
- Support Streaming Output: on
- Support Plugins: on
- Support Image Input: on
- Support PDF Input: off
- Support External Image URLs: off
- Support Thinking: off
- Support Prompt Caching: off
Click Test & Save. If your browser asks to let www.typingmind.com access other apps and services on this device, click Allow. This lets the page reach Ollama at localhost.
Chat with Llama4 in TypingMind
Pick Llama4 16X17B (local) in the model menu under the chat box and say hi. The first reply can take longer while Ollama loads the model into memory. After that, replies come straight from your own computer.
Llama4 16X17B (local)Use plugins with Llama4
Llama4 supports tool calling, so it can use TypingMind plugins. Open the plugin menu next to the model picker and turn on the ones you want, like web search, image generation, or Mermaid diagrams. The model calls them when it needs them.
Llama4 16X17B (local)Send images to Llama4
Llama4 can read images. Click the paperclip under the chat box, attach a photo or screenshot, and ask about it. The image stays on your computer.
You can now use your private AI with TypingMind's chat features: prompts, AI agents, projects, and switching between your local model and cloud models in the same chat.
Tips and troubleshooting
1. The first message of a new chat seems stuck
TypingMind writes a title for each new chat. If the title is made by the same local model, Ollama gets two requests at once and answers them one at a time, so your reply waits behind the title. Go to Settings → Internal prompts, open Advanced Settings under Auto generate title for new chats, and pick another model for title generation, or turn auto titles off.
2. "Could not connect to the API (code: 1)"
TypingMind can't reach Ollama. Check these in order:
- Ollama is running (you see its icon in the menu bar or taskbar).
- You set
OLLAMA_ORIGINSand restarted Ollama afterwards. - You clicked Allow when the browser asked about accessing apps on this device. If you clicked Block, allow it again from the site settings icon next to the address bar.
- You're using TypingMind in the browser, not the Mac or Setapp app.
To check that Ollama is answering, run:
curl http://localhost:11434/api/version3. See what Ollama is doing
Ollama logs every request with its status code (200 means it worked).
- macOS:
tail -f ~/.ollama/logs/server.log - Linux:
journalctl -u ollama -f - Windows: open
%LOCALAPPDATA%\Ollamaand readserver.log
4. Replies are very slow or Ollama runs out of memory
The model is probably too big for your computer. Pick a smaller size from the size table, or close other heavy apps. As a rough rule, you need about the download size plus 2 GB free memory.
Frequently Asked Questions
Is Llama4 free to run locally?
Yes. Llama4 is an open model you download once to your computer. Every message runs on your own machine, so there is no per-message cost and no API key to manage.
Is Llama4 private when I run it locally?
Yes. Your messages go from your browser to the local AI tool on your own computer, not to an AI company. TypingMind keeps your chat history in your browser. If you turn on Cloud Sync, your chats are also backed up to TypingMind's cloud.
How much RAM do I need for Llama4?
The smallest size, llama4:16x17b, is a 67 GB download and runs comfortably with about 96 GB of RAM. The default size, llama4:16x17b, needs about 96 GB. Bigger sizes are smarter but need more memory.
Does Llama4 work offline?
The model runs fully offline once it's downloaded. The TypingMind web app still needs to load in your browser.
Can Llama4 use TypingMind plugins?
Yes. Llama4 supports tool calling, so turn on Support Plugins and it can use plugins like web search. Small sizes may call tools less reliably than large ones.
Can Llama4 read images?
Yes. Llama4 accepts images, so turn on Support Image Input and attach images in the chat.
Why is the first reply slow?
The local AI tool loads the model into memory on the first message and keeps it there for a few minutes. Later replies start faster.
Does TypingMind run locally?
TypingMind runs in your browser at www.typingmind.com, and your chats are stored locally in your browser by default. When you chat with a self-hosted model, your messages go straight from your browser to the model on your own computer, not through TypingMind’s servers. If you turn on Cloud Sync, your chats are also backed up to TypingMind’s cloud.
Can I self-host TypingMind?
Yes. To self-host TypingMind for your team or company, use the self-hosted version of TypingMind Teams. You run it on your own servers next to your self-hosted models, so chats, users, and settings stay inside your own infrastructure. Contact our sales team to get started.
