Odysseus

Odysseus is a free AI workspace that you install on your own computer. Instead of paying for ChatGPT or Claude and sending your words to their servers, you get one app that does chat, research, email, notes, documents and more, using AI models that run on your own machine. YouTuber Felix Kjellberg, better known as PewDiePie, built it for himself and released it on GitHub on 31 May 2026. Within weeks it had tens of thousands of stars, and it's now looked after by a wider group of volunteers under the odysseus-dev name.

What it can do

• Chat and agents. Talk to an AI model like you would in ChatGPT, or hand it a task and let it work through the steps itself. Agents can use tools such as web search, files and the command line, connect to MCP servers, and remember things between chats.
• Cookbook picks models for you. Cookbook looks at your computer's graphics card and memory, then recommends models that will actually run on it, from a catalogue of more than 270. It downloads and starts them for you.
• Deep research. Ask a question and Odysseus searches the web, reads the sources and writes you a report.
• Compare models blind. Send one question to several models and see their answers side by side without knowing which is which, so you can pick the best one for your job.
• Email that sorts itself. Connect your inbox and it tags messages, summarises long threads, catches spam, sets reminders and drafts replies for you to check.
• Documents, notes, tasks and a calendar. A writing editor where the AI can suggest edits, plus notes, to-do lists, reminders and a calendar that syncs with the one on your phone.
• Private by default. It runs on your own computer with no tracking. You can add paid online models from companies like OpenAI or Anthropic if you want, but you don't have to.

Where you'll meet it

Odysseus doesn't run AI models by itself. It sits on top of engines like Ollama and llama.cpp, or LM Studio, and adds the workspace around them. It also comes with Chroma to store its memory and SearXNG for private web searches. PewDiePie is also working on Ajax, a small AI model made to run inside Odysseus, but it isn't available to download yet.

Odysseus, Open WebUI or AnythingLLM?

All three give AI models on your own computer a friendly window like ChatGPT's. They differ in how much they do beyond chat and in their licences.

OdysseusThis page

Best forAn all-in-one private workspace

Chat plus agents, research, email, notes, calendar and documents in one app, with Cookbook to pick models for your computer. Very new, so expect rough edges and frequent changes.

How you use it

In your web browser, after installing it on your computer

Open source

Yes (AGPL 3.0)

Cost

Free

Open WebUI

Best forA polished chat app for a team or family

The most popular chat window for local models, with user accounts, document uploads and lots of add-ons. Focused on chat rather than email or notes. Its licence asks you to keep its branding.

How you use it

In your web browser, after installing it on your computer or a server

Open source

Source available (Open WebUI licence)

Cost

Free; paid plans for businesses

AAnythingLLM

Best forChatting with your own documents

A simple desktop app built around asking questions about your files, with agents and a choice of local or online models. The easiest of the three to install.

How you use it

A desktop app, or in your browser on a server

Open source

Yes (MIT)

Cost

Free; paid cloud hosting

A quick way to choose: want one private app to replace ChatGPT, your notes app and your inbox helper? Try Odysseus. Setting up a chat app for several people? Open WebUI is the safer bet. Mainly want to ask questions about your own documents with the least setup? Pick AnythingLLM.

Good to know

• Installing it takes a few typed commands. The easiest way is Docker; there's also a one-command launcher for Windows and a start script for Macs. Early reviewers found the setup tricky, so give yourself an hour.
• The app itself is light, but running AI models on your computer needs a good graphics card and plenty of memory. On a weaker computer, connect it to an online model or another computer instead.
• On an Apple Silicon Mac, run it directly rather than in Docker, or it can't use the Mac's graphics chip.
• It's only a few months old and changes every day. The main branch is the steadier choice.
• Keep the login switched on and don't open it to the internet, because its agents can run commands on your computer.
• Free models on your own computer are good for everyday jobs but not yet as smart as the best paid assistants for hard tasks.
• Licence: AGPL 3.0. Free to use, change and run for yourself or your business. If you change it and let other people use it over the internet, you must share your changed code under the same licence, so it doesn't suit selling a closed, paid version.