If you searched for PewDiePie's AI repository, this is the project you are looking for: odysseus-dev/odysseus on GitHub. Odysseus is a free, self-hosted AI workspace for chat, agents, research, documents, memory, email, calendar tools, and local or remote language models.
The repository has changed since its early launch. It now lives under the Odysseus organization rather than the original PewDiePie-branded location. It is also moving quickly: the default dev branch has the newest work, while the maintainers describe main as the more stable, curated branch.
This guide shows the official installation routes for Docker, Windows, Linux, and Apple Silicon. I will also show you how to connect Ollama, find the first-login password, and avoid the security mistake that matters most: exposing an AI admin console directly to the public internet.
Official links
What you are installing
Odysseus is the workspace, not the AI model itself. The application can connect to cloud APIs, an Ollama server, or other OpenAI-compatible model endpoints. It also includes Cookbook features for downloading and serving models.
This distinction matters for hardware. The Odysseus app is relatively lightweight. Local model inference is the part that consumes RAM and GPU memory. You can install the workspace on a modest computer and connect it to an API or a model server elsewhere. You do not need PewDiePie's multi-GPU workstation just to try the project.
If hardware is your main concern, read my separate guide: How to Build a Local AI Workspace Like PewDiePie's Odysseus: Hardware, Models, and Cost.
Before you start
Pick one installation route:
Docker: the recommended and most repeatable setup for most users.
Native Windows: convenient if you already use Windows and want to connect to Ollama.
Native Linux: useful when you want direct access to the host and GPU stack.
Native Apple Silicon: the better choice for Metal-accelerated local models on an M-series Mac.
You need Git. Native installations also need Python 3.11 or newer. Docker users need Docker with Compose support. The repository's dependencies are not pinned, so installs performed months apart can resolve to different package versions. If this is a serious deployment, keep notes about the commit and dependency versions that worked for you.
Option 1: install Odysseus with Docker
The maintainers recommend Docker for the quickest start. These commands follow the official setup guide:
git clone https://github.com/odysseus-dev/odysseus.git
cd odysseus
cp .env.example .env
docker compose up -d --build
The copied .env file is optional, but it makes your deployment settings explicit. When the containers become healthy, open:
http://localhost:7000
Odysseus creates an admin account during first setup and prints a temporary password. For Docker, retrieve it from the container logs:
docker compose logs odysseus
Sign in with the generated password, then change it in Settings. If port 7000 is already in use, set APP_PORT=7001 in .env, recreate the container, and open port 7001 instead.
Docker Compose binds the web interface to 127.0.0.1 by default. That is a good default. Do not change APP_BIND to 0.0.0.0 unless you understand why you need network access and have authentication plus a trusted private-access layer in place.
Stable branch or development branch?
A normal clone currently checks out dev, the default branch. It contains the latest changes but may be unstable. If you prefer the curated branch, clone main explicitly:
git clone --branch main https://github.com/odysseus-dev/odysseus.git
For a first installation, I would choose main unless you need a feature or fix that only exists on dev.
Option 2: install Odysseus natively on Windows
Windows has a one-command PowerShell launcher that creates the virtual environment, installs dependencies, runs setup, and starts the server:
git clone https://github.com/odysseus-dev/odysseus.git
cd odysseus
powershell -ExecutionPolicy Bypass -File .\launch-windows.ps1
Open http://localhost:7000 and use the temporary admin password printed in PowerShell.
If you prefer to run each step yourself:
git clone https://github.com/odysseus-dev/odysseus.git
cd odysseus
py -3.11 -m venv venv
venv\Scripts\Activate.ps1
pip install -r requirements.txt
python setup.py
python -m uvicorn app:app --host 127.0.0.1 --port 7000
If Python 3.11 is unavailable but you have a newer supported version, replace py -3.11 with the installed version, such as py -3.12.
The core workspace runs natively on Windows. For Cookbook background downloads and the agent shell tool, install Git for Windows so bash.exe is available. Native vLLM and SGLang model serving still belongs on Linux or WSL2. For most Windows users, Ollama is the simpler local-model route.
Connect Odysseus to Ollama on Windows
Install Ollama, start it, and download a model that fits your computer. Then add this OpenAI-compatible endpoint in Odysseus Settings:
http://localhost:11434/v1
If Odysseus runs in Docker while Ollama runs directly on the host, use:
http://host.docker.internal:11434/v1
Ollama must listen somewhere the container can reach. The official guide uses:
OLLAMA_HOST=0.0.0.0:11434 ollama serve
Do not expose Ollama's port to the public internet. This address is for communication between your host and container, not an invitation to publish port 11434 through your router.
Option 3: install natively on Linux
Linux and macOS share the manual Python route:
git clone https://github.com/odysseus-dev/odysseus.git
cd odysseus
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python setup.py
python -m uvicorn app:app --host 127.0.0.1 --port 7000
You need Python 3.11 or newer. Cookbook also uses tmux for background model downloads and servers. Open http://localhost:7000 after startup.
For NVIDIA GPUs in Docker, Odysseus includes a diagnostic script:
scripts/check-docker-gpu.sh
The default mode only diagnoses the setup. It does not install packages or edit .env. The official guide also documents assisted NVIDIA Container Toolkit setup and separate AMD/ROCm overlays. Read those instructions before enabling GPU passthrough because a working nvidia-smi inside the container does not guarantee that llama.cpp was built with CUDA support.
Option 4: install on an Apple Silicon Mac
Docker cannot give the container access to Apple's Metal GPU. If you want GPU-accelerated local inference on an M-series Mac, use the native launcher:
git clone https://github.com/odysseus-dev/odysseus.git
cd odysseus
./start-macos.sh
The macOS script starts Odysseus at:
http://127.0.0.1:7860
Port 7860 is intentional because AirPlay Receiver commonly occupies port 7000 on macOS. The script installs Homebrew dependencies, creates the Python environment, runs setup, and starts the server.
Odysseus can use llama.cpp or Ollama with Metal on macOS. vLLM and SGLang target CUDA or ROCm and do not run natively on macOS.
What model should you try first?
Start smaller than your ambition. On an 8GB laptop GPU, the maintainers recommend beginning with a GGUF model using Q4 quantization and llama.cpp. That is easier to validate than jumping straight into GPTQ or AWQ models served through vLLM or SGLang.
A machine without a dedicated GPU can still use Odysseus. Connect it to a cloud API, a remote model server, or run a small model on the CPU. Model memory depends on the quantized weights, context length, KV cache, runtime workspace, and available headroom. A parameter count alone does not tell you whether a model will fit.
First checks when something fails
For Docker, these commands answer most first questions:
docker compose ps
docker compose logs --tail=120 odysseus
Check that the container is healthy, confirm which port is bound, and look for the generated admin password. If the browser cannot connect, make sure you are opening the host port rather than an internal service port.
If Ollama works on the host but Odysseus cannot see it from Docker, check the endpoint, confirm Ollama is listening beyond its own loopback interface, and use host.docker.internal instead of localhost inside the container configuration.
On Windows, an .env file saved with a UTF-8 byte-order mark can make the first setting behave unexpectedly. The current project includes handling for this case, but saving as UTF-8 without BOM remains the safest choice.
Do not expose Odysseus like a normal public website
Odysseus can run shell commands, read and write files, manage models, access email and calendars, and store API tokens. The project's threat model says to treat it like an admin console. It is designed for trusted users on a private network, not anonymous visitors.
Keep AUTH_ENABLED=true. Keep internal services such as ChromaDB, SearXNG, Ollama, vLLM, and llama.cpp private. If you need access away from home, use HTTPS behind a trusted reverse proxy or private layer such as Tailscale, a VPN, or Cloudflare Access. Do not forward Odysseus or raw model-server ports directly from your router to the internet.
Self-hosting also does not automatically make every interaction private. If you configure cloud model APIs, web search, email, or other remote services, data can still leave your computer through those integrations.
Is Odysseus worth installing?
Yes, if you want one self-hosted place to experiment with models, agents, documents, memory, and personal integrations. The project is unusually ambitious, and that comes with a tradeoff: it changes quickly and gives administrators powerful capabilities. Expect to read logs and configuration notes occasionally.
If your only goal is chatting with one local model, Ollama plus a simpler interface may be easier. If you want a broader AI workspace that you can inspect, modify, and connect to your own services, Odysseus is worth trying. Use your existing computer first. Buying a pile of GPUs before you know which workflows matter is the expensive way to learn.
For the background story and a closer look at how the project has evolved, read What PewDiePie Is Building in AI Now.
References
Originally published at https://blog.jenuel.dev/blog/pewdiepie-ai-repo-install-odysseus-locally
Thanks for reading! If you enjoyed this article and like this kind of content, you're always welcome to buy me a little coffee, but only if you'd like to. No pressure at all, and either way I'm truly grateful you stopped by. ☕️
Top comments (0)