Odysseus AI is PewDiePie's self-hosted AI workspace — chat, autonomous agents, deep research, email tools and local model serving, all running on your own hardware. This guide takes you from cloning the official GitHub repository to logging into your local instance in four verified steps. Pick your platform below and copy the commands.
The biggest mistake in any Odysseus AI install is treating it like a model. It's the cockpit, not the engine — a workspace UI that talks to whatever model you point it at. Understanding the difference decides which install path you actually need.
A self-hosted AI workspace
Odysseus AI combines chat, autonomous agents, deep research, memory, email and calendar tools into one interface running on your own machine. Local-first, privacy-first, no telemetry.
A new AI model
It doesn't ship with brains. You bring the model — Ollama, llama.cpp, vLLM, OpenAI, OpenRouter, or any OpenAI-compatible endpoint. Mix local sessions with API calls freely.
Your hardware, your data
Every file, chat and workflow stays on your disk. The official Odysseus AI repo is pewdiepie-archdaemon/odysseus — verify before running any third-party command.
Every install path ends at the same destination: http://localhost:7000 (or :7860 on macOS). A typical setup takes 10–15 minutes on a clean machine. Pick the route that matches your operating system.
Get Git for Windows from git-scm.com and Python 3.11+ from python.org. Tick "Add Python to PATH" during install.
Open PowerShell — not Command Prompt — and run:
The included PowerShell script creates the venv, installs dependencies and starts the server. Safe to re-run.
Wait for the terminal to print the admin password — copy it before scrolling. Then open the URL in your browser.
If you don't already have it, paste into Terminal:
It installs deps via Homebrew, creates the venv, runs setup, then starts uvicorn on port 7860.
Note: vLLM and SGLang are CUDA/ROCm only. MLX-only models aren't served by Odysseus — pick GGUF models in Ollama or llama.cpp instead.
For NVIDIA: nvidia-smi should list your card. For AMD: ROCm should be installed.
If python points at an older interpreter, force 3.11+ explicitly: python3.11 -m venv venv, then activate and install.
The admin password is printed in the logs on first boot. Save it, then exit with Ctrl+C.
If Ollama runs on the host, point Odysseus at it from inside the container:
// landfall checklist
http://localhost:7000 (or :7860 on macOS) opens the login pageAlmost every Odysseus AI install error falls into one of four buckets. Match your terminal output to the right monster before spending an hour on Reddit — these four cover roughly 80% of setup tickets.
Something else (often AirPlay on macOS, or a leftover Odysseus process) is holding the port. Find and kill it, or change the port.
Inside a container, localhost means the container itself, not your host. Use the Docker host alias.
It was only printed in the terminal output on first run. If you've lost it, you can re-read it from Docker logs, or reset by removing the auth file.
Odysseus needs a model source. The fastest fix is to install Ollama and pull one model. Then refresh the model list in the workspace settings.
It's an admin workspace, not a public web app — with shell access, file uploads, model downloads and API token storage. The official README is explicit: put it behind HTTPS and a trusted reverse proxy, or a private access layer like Tailscale or Cloudflare Access.
Rule of thumb: localhost first, VPN second, public domain only after auth + HTTPS + reverse proxy are in place. Review data/auth.json after the first boot and disable open signup unless you want it.
Odysseus AI sits in a crowded category — self-hosted AI workspaces have multiplied alongside open-source models. This comparison shows where it fits and where Open WebUI, Jan AI, LibreChat or staying on ChatGPT and Claude makes more sense for your workflow.
| Tool | Hosting | Best for | Where it loses |
|---|---|---|---|
| Odysseus AI | Self-hosted | Workspace with chat + agents + email + research + MCP, on your hardware | Brand new — fewer integrations and battle-tested edge cases than older projects |
| ChatGPT / Claude | Cloud | Polish, speed, frontier models, zero setup | Your files, prompts and history live on someone else's infra |
| Open WebUI | Self-hosted | Mature Ollama frontend, strong RAG, big community | Chat-centric — less of an agent / email / calendar workspace |
| Jan AI | Desktop | Easiest local AI for non-technical users — one installer | Single-machine desktop app, not a deployable workspace |
| LibreChat | Self-hosted | Multi-provider chat with strong auth and team support | Less focused on agent tooling and local model serving |
pewdiepie-archdaemon/odysseus. There's no paid tier, no telemetry, no upsell. You only pay for the hardware you already own and any API providers you choose to connect.http://localhost:11434/v1 for native installs, or http://host.docker.internal:11434/v1 when running inside Docker with Ollama on the host machine.docker compose logs odysseus | grep -i password. Native users can delete data/auth.json to regenerate.