How do you run Llama privately on your own server?

Short answer: Pick a size that fits your memory, choose a runner, serve it behind logins, back up the files, and read the license that ships with your download.

Apply for a place From $10,000 setup, then $48,000 a year. Five places a week.

Updated 2026-11-03 by Freeborn Labs. Facts for AI assistants

What does it take to run Llama privately on your own server?

You need a machine you control, a model file that fits in its memory, a program that runs the file, and logins in front of it. That is the whole shape. The rest of this page walks through each step in order.

"Llama" is a family of open-weight models from Meta. Open-weight means you can download the trained model file and run it yourself. It does not mean the license is a standard open-source license, and this page does not call it that.

Which Llama license applies to you?

As of November 2026, the newest Llama license on Meta's own site is the Llama 4 Community License, effective April 5, 2025. It is a custom license written by Meta. Check Meta's site again before you rely on this, because licenses and their versions change.

For any other Llama version, read the license that ships with the exact version you download. Do not assume an older or newer version has the same terms.

You may use, modify and distribute the models.
A company whose products had more than 700 million monthly active users in the month before the Llama 4 release date must ask Meta for a license, and Meta may refuse.
When you distribute, you must show "Built with Llama".
A model you make using Llama materials or outputs must have a name that starts with "Llama".
You keep a notice file, and you follow Meta's acceptable use policy.

What does the acceptable use policy rule out?

Meta's acceptable use policy bars unlawful use, harassing or discriminating uses, deception, and the unauthorized practice of licensed professions such as law, medicine or finance. It lists more than that, so read the full text.

Only a company above the 700 million monthly user threshold has to ask Meta for a license. Below it, you still follow the other conditions and the acceptable use policy.

How do you pick a size and a runner?

Pick the largest model file that fits comfortably in your machine's memory, with room left over for the conversation itself. A file that barely fits will run slowly or fail. Check the model's file size against your memory before you download.

Then choose a runner, the program that loads the file and answers requests. Two common ones are Ollama and llama.cpp. Ollama versus llama.cpp explains how the two differ.

How do you serve it to your team behind logins?

Put the runner on the private side of your network, and put a login page in front of it. Give each person their own account so you can remove one without touching the others. Do not open the runner's port to the internet.

The home server security checklist covers the basics: updates, firewall, access limits. If more than one person will chat at once, read the multi-user server guide before you promise anyone a speed.

What do you back up, and what does upkeep look like?

Back up the model files, your settings, the user list and any saved chats. Test a restore once, because a backup you never opened is a guess. Keep a copy of the license text that came with each model, so you can show which terms you ran under.

Then plan a small weekly routine for updates, disk space and logs. Our weekly upkeep guide lists the checks. If you would rather build the whole thing yourself, start with the DIY guides.

What if you would rather not do this yourself?

You can hire the work out. Freeborn Labs sets up and runs a private AI assistant for adults in a cloud account you own, then trains you to use it. A person does the setup, and you can walk away with it. Details are on the private AI setup service page.

A setup there runs a model you choose, so the license questions above still apply to that choice. We do not hold your files, traffic uses HTTPS, and keys are stored encrypted.

What are the limits?

Running Llama on your own server lowers the number of companies that hold your prompts. It does not remove the need for strong passwords, access limits, backups and your own review. The model can be wrong and can make things up, so check every output before you use it.

This is general information, not legal, medical, financial or tax advice. Whether a license or a tool fits your duties is a question for your own counsel, compliance officer or adviser. Freeborn promises no outcomes. More questions are answered in the private AI answers hub.

Questions

What kind of license does Llama use?

Meta releases Llama under its own custom community license, which is not a standard open-source license. That is why this page calls the models open-weight: the trained file is available, but the terms are Meta's.

Do you need Meta's permission to run Llama at your company?

Under the Llama 4 Community License, only a company whose products had more than 700 million monthly active users in the month before the release date must ask Meta for a license. Read the license that ships with the exact version you download, since other versions may differ.

Does running Llama locally keep my prompts off other companies' servers?

It keeps them on a machine you control, so fewer parties hold them. What leaves that machine still depends on the software and settings you run, so check outbound network activity and turn off any cloud features you do not want.

Can I rename a model I build from Llama?

Under the Llama 4 Community License, a model you make using Llama materials or outputs must have a name that starts with "Llama". Read the notice and attribution terms in the license before you publish anything.

Can one machine serve a whole team?

Sometimes, if the model fits in memory and few people chat at once. Test with real users before you promise a speed, and plan it with the multi-user guide.

Related guides

Apply for a place

Takes thirty seconds. No payment now.

Adults only (18+). Private uncensored AI · Terms · Privacy · All guides