What is Odysseus? Free Artificial Intelligence of PewDiePie

A video published with this title announced the last ring of a project documented parts for months. PewDiePie, namely Felix Kjellberg, released Odysseus for free artificial intelligence platform running on its own hardware. This open source tool that offers alternative services such as ChatGPT and Claude without sending data to the cloud; from chat to autonomous agents, collect a wide range in one interface. In this article, what is Odysseus, which properties offer and how it was established.

What is Odysseus and What Does it Work?

Odysseus is an artificial intelligence workspace that you host on their server to work with language models. Project's official GitHub store The platform is defined as a tool that collects chat, agents, research, documents, email, notes, calendar and local model workflows in one place. Introduction page If the approach is downloaded to three principles: there is no local- priority, privacy-previous and telemetry.

The intention of expression in the official definition clearly shows:

"Local- Priority, privacy-previty and no telemetry. Only you and models. "

The purpose of the platform is to mimic the ease of use of popular services without delivering data to third-party servers. The user is connected to local models in its own machine, or prefers an external API. The decision was completely abandoned to the user. ChatGPT and other language models that are previously compared to different artificial intelligence models are a logic that is the opposite of the only-centered cloud approach that stands out in the discussion.

How did PewDiePie Improve Artificial Intelligence Project?

Odysseus did not appear overnight. The foundation of the project is based on a personal system called ChatOS before. According to the news published in Tom's Hardware The hardware set by Felix consists of 2 RTX 4000 Adas and 8 mode RTX 4090 screen cards with 48 GB memory. TheOutpost is transferred by This installation of about 20,000 dollars was converted to a mini data center of 10 GPUs with PCIe bifurcation technique and can process context window up to 100,000 tokens.

Open source models were run like Llama 70B and Qwen in the system. According to Tom's Hardware, Felix says "ChatGPT, but much faster", OpenAI publishes GPT-OSS open source language model also included among models. It was possible with techniques such as running models in a local employee installation, quanification and tensor parallelity.

The most interesting part of the work was that models were spoken with each other. According to the news published in Dexerto Felix first founded a structure called "The Council". Multiple bots respond to the same question, then selected with the best answer vote. The weakest boots were removed without being addressed. After construction, it became "The Swarm" expanding to 64 bot.

Then it was something strange. Bots began to lose the answer of each other to not be denied. The system was taken into a simple structure when it was noticed this cooperation. A moment that attracts attention in the process was the self-development of the model. According to Tom’s Hardware’s transfer, Felix was printed the code again to add new features to ChatOS. "Machine is doing the machine" said at this stage. When internet access is provided to the model, the response becomes markedly hit, with the addition of RAG, the deep research ability was gained.

Felix’s overall orientation in this period was also remarkable: to purify their life from Google services, learn from scratch and donate the GPU power in idle Folding@home disease research. It was also mentioned in the planning to train its own model in previous descriptions. Odysseus appeared as a mature state of this trial. From a hobby project to a platform where everyone can download and install.

What Features Does Odysseus Offer?

Odysseus also accommodates different modules under which it looks like a single application. The components that stand out according to the description in the official warehouse are listed as:

Chat: connects to any local model or API. vLLM, llama. cpp, Ollama, OpenRouter and OpenAI options are supported; SGLang in the installation guide and other end spots compatible with OpenAI are also passing.

Agent: This module built on opencode infrastructure; MCP, web, file, shell, skills and memory tools to run one task from beginning to finish.

Cookbook: The hardware is combing, it suggests models that are suitable for him and is running with one click. Technical information is a feature that reduces installation threshold for limited users.

Research: Deep research tasks are executed. A topic is sought, it is immersed in the relevant information from there.

Model offering: Local models are served directly through the platform.

Comparison: The answers to multiple models to the same question are side-by-side, if you want to see what model is written (kör test).

Documents: You are writing text, it helps with artificial intelligence editing and recommendation. Markdown is supported by HTML and CSV.

Email: The box connected with IMAP/SMTP produces priority, labeling, summary, reminder and response draft.

Notes, tasks and calendar: Recalled notes, the list of things to do is offered to sync the calendar with scheduled tasks and CalDAV when the agent can run.

These features were shown only in the promotional video. The interface’s research tasks were transferred to run, manage conversations and work like a special assistant without addiction to external services.

Why is Odysseus in the Center?

The most prominent side of the platform is privacy. Chats, files and personal data are kept in the hardware that the user controls; they are not sent to external servers. This approach, which is subject to local-previty (local-first) passes into account the delivery of data to large-tech companies.

The logic of Felix is based on the following: the model becomes so useful if it learns a lot of contexts about the user. The price of this context in cloud services is out of the data. The same context in the local installation is obtained without any exit. The difference here. Odysseus's introduction page The philosophy of the project is summarized in a single sentence:

"No sales team, no demo request, no Trojan horse. "

It is necessary to read correctly; here implies, for closed systems that collect data in the background of subscription traps and backdrops. Not telemetry means that the platform does not even collect usage data.

How does Odysseus separate from ChatGPT and Claude?

Odysseus aims at the same user experience with ChatGPT and Claude. The main difference arises where the work works and where the data remains. Odysseus with the intimate expression in the official definition is positioned as a version of these services "a little more cashier and fun". The following table compares three approaches with basic headers:

Feature	Home	News Home	Home
Home	Local / self-hosted	Cloud	Cloud
Home Tags	Free, open source	Free + paid plan	Free + paid plan
Data control	User's hardware	Provider server	Provider server
Telemetry	No	Yes	Yes
Model Selection Home	Local model or API	Your models Home	Your models Home

The table shows the flexibility of Odysseus. The user can connect OpenRouter or OpenAI API and work with cloud models. So the preference does not be beaten. This corresponds to ChatGPT and Claude; ease of use, stability and direct support. They do not require installation, they are used on the browser. Odysseus wants some technical labor. The difference between a corporate service with an open source project growing with voluntary contributions is manifested here.

Is Odysseus Free and How to Install?

Odysseus free and open source; is published by AGPL-3.0-or-later license. You can download and modify the project and host it on your own server. Official warehouse Home quick start steps as follows:

1. Copy the storage: git clone https://github.com/odysseus-dev/odysseus.git

2. Enter Folder: cd odysseus

3. Create set file: cp .env.example .env

Start with Docker: docker compose up -d --build

Open http://localhost:7011 in the browser when containers are ready. The first admin password writes docker compose logs in the output of audiosseus.

Once you have logged in, you redirect Odysseus from the settings panel to the model you want to use. Local installation options with Python 3.11+ for Linux, macOS and Windows if you want Docker Home installation guide tell. The giant with default branch takes the latest changes before; you can use the main branch if you want a more seated version.

This accessibility distinguishes project from other local AI trials. While many open source tools appear like a developer debugging panel, Odysseus aims a modern assistant interface. A similar privacy and cost advantage, previously compiled free artificial intelligence tools There were a few products that offer a local model option in the list. Odysseus expanded this approach under one roof.

A part of the project according to the Lansman version (v1.0) README file was developed via the phone, with the installation of Termux shells and PWA. This note was then removed from README; the mobile compatible interface (PWA) was included in the launch feature list.

What Changed from June 2026?

Odysseus quickly improved after release as v1.05.2026. Important changes from 14.10.2026:

• New address: From the warehouse pewdiepie-archdaemon account odysseus-dev/odysseus moved to the organization. The old address leads to new warehouse.

• Installation: Quick start is done via Docker Compose and the interface is opened in 7011 port. The official multi-arched Docker image is published on GitHub Container Registry; the last version tag is 1.0.3.

• Platforms: Linux, macOS (Metal supported llama in the installation guide. with cpp/Ollama) and local installation ways for Windows are explained.

Branches: The default branch takes the latest changes to the giant; the main contains more selected changes.

Community: The warehouse has reached over 93.000 stars; the account that contributes more than 300 contributors on GitHub.

Road map: Road map priority debugging is ranked as clean installation tests in different systems and reliable operation in different hardwares of Cookbook.

Ajax: Model for Odysseus

PewDiePie announced a model called Ajax in October 2026. Ajax, open source of Alibaba Qwen 3.5 9B is prepared by fine tuning and designed for web search, navigation, email and calendar works in Odysseus. According to Decrypt's news dated 03.10.2026, the model has not yet been downloaded. An official model card or license information could not be verified.

What is the Community

Lansman quickly spread in the software and AI community. Responses are collected in two main groups.

Supporting: Evaluation of SquaredTech It was noted that the project was not appreciated by itself. Most open sources have highlighted that the AI project itself is big, that Odysseus is a tool that knows what it is. It is seen as a strong starting point for those who want a modern assistant without delivering the data to the cloud. As high as the demand for local vehicles is not at all, it feeds attention.

Critical: There is a question that is taken into consideration in the same rating: a 1.0 version that publishes a single developer can protect the large set of features that it is lanized? E-mail clients, calendars, agent frames and vector memory systems are serious engineering fields alone. A hard job to maintain all at the same time. The suspicions of stability and long-term care are collected at this point.

Two sides also comment differently the same fact: Odysseus is an ambitious start. The more sustainable this claim will be seen over time.

Frequently Asked Questions

What is Odysseus?

Odysseus is a free and open source artificial intelligence working area that PewDiePie published in 31.05.2026. You host in your own hardware; chat, agents, research, documents, email, notes and calendar tools collect in one interface. ChatGPT and Claude offers a local and privacy-oriented alternative to services.

Is Odysseus free?

Yes, Odysseus is totally free and open source. It can be downloaded and replaced from the official warehouse and is hosted on the user’s own server. There is no sales process, demo request or subscription fee.

How to install Odysseus?

Recommended way Docker. Copy the warehouse with git clone and enter the folder, you copied .env.example file as .env and run docker compose up -d-build command. Then you open http://localhost:7011 in the browser. If you don’t want to use Docker there are local setup steps for Linux, macOS and Windows in the installation guide.

Do you need the hardware of PewDiePie to use Odysseus?

No. Odysseus can also connect to external APIs such as OpenRouter or OpenAI, as user can work with local models on their computer. The 8-screen card installation of PewDiePie is only its personal system; the platform can work in much more modest hardware.

Does Odysseus collect my data?

According to the official recognition, Odysseus does not contain telemetry, meaning the usage data is collected. Chats and files are kept in the hardware controlled by the user and are not sent to external servers.

Does Odysseus ChatGPT and Claude replace?

Odysseus is aimed at a similar experience of use but a young open source project that is maintained with voluntary contributions. ChatGPT and Claude stability are leading in terms of ease of use and direct support. If the privacy and data control is priority, Odysseus is a strong preference; cloud services remain more convenient if it calls uninterrupted corporate support.

What models do Odysseus work?

Odysseus supports local model run tools such as vLLM, llama.cpp and Ollama and external APIs such as OpenRouter. In the previous trials of PewDiePie, open source models such as Llama 70B, GPT-OSS and Qwen were used. In October 2026, he announced a model called Ajax, which was fine-tuned for Odysseus; according to the news released in the press 03.10.2026 was not yet downloaded.

Odysseus is an intermediate of PewDiePie’s trials. Starting from ChatOS, the journey from the boat army that voted is connected to a warehouse where everyone can download and updated regularly since the warehouse launch. This is not yet clear how much a large set of features will be stable in the long run. If you want to meet other developers who are trying to install their own artificial intelligence in Masqot communities You can browse.