About Llama.Online

Llama.Online is a free website for chatting with Meta's Llama models. You open the page, type a question and get an answer from a real Llama model, with no account, no download and no payment. This page explains who we are, what runs behind the chat and how the site pays for itself.

Who we are

We are a small independent team that builds and maintains a handful of free AI chat websites. We are not part of Meta, and Meta does not run, sponsor or review this site. We started Llama.Online because Llama is one of the best-known open-weight model families, yet trying it usually means installing software, renting a server or signing up for a developer platform. We wanted a version that anyone could open on an ordinary phone and use straight away.

The Llama models in our chat

The chat offers four official Meta releases. You pick one in the model menu, and your choice is remembered for each conversation:

• Llama 4 Maverick is the default. It is a Llama 4 mixture-of-experts model that reads text and images and handles most everyday tasks well.
• Llama 4 Scout also reads images and accepts the longest inputs of the four, which helps with long documents.
• Llama 3.3 70B is a text-only model from December 2024 and a strong all-rounder for writing and reasoning.
• Llama 3.1 8B is the smallest and quickest option, good for short and simple requests.

We do not train or change these models; a cloud inference provider runs them for us. Meta's later models, released under the Muse name, are not part of our chat. For the background, read what Llama is or follow the history of the Llama family.

How the chat works

When you send a message, your browser passes it to our server. The server adds a short set of instructions (for example, the assistant you picked), checks your daily allowance and forwards the request to our AI gateway, which sends it on to the inference provider. The answer streams back word by word, so you can start reading before it is finished. The models cannot browse the internet and their knowledge ends at a training cut-off, so they do not know recent news.

Around the chat we built the parts that make it useful day to day: conversation history with search, pins and export, sixteen assistants, a prompt library, twelve single-purpose AI tools, share links and an interface in 26 languages.

Free, and paid for by ads

Running large language models costs money for every message. We cover that cost with advertising shown around the content of our pages. Ads never appear inside the model's answers, and advertisers have no say in what the chat writes. To keep the service sustainable and to prevent abuse, each visitor gets a daily allowance of free messages, currently 100 a day, shared between the chat and the tool pages. It resets every day.

Your privacy in brief

You do not need an account, and your conversations are saved only in your own browser. We do not store the text of your messages on our server; it passes through to produce the answer. Our gateway keeps short technical logs, such as which model was used, the size of the request and a hashed form of your IP address. If you share a conversation, a read-only copy is stored on our server until you ask us to remove it. The privacy policy has the full details, including how ads and analytics work.

Independent from Meta

Llama is a trademark of Meta Platforms, Inc. We use the name only to describe the models that answer in our chat. Meta publishes Llama under its own community license, which asks products built on the models to say so. That is why you see "Built with Llama" on every page. A plain-language summary of the license is on our Llama license page.

Get in touch

Found a bug, want a shared chat removed or have an idea for the site? Write to us through the contact form, and add your e-mail address if you would like a reply. Answers to common questions are on the FAQ page.