Odysseus

Odysseus is a self-hosted AI workspace created by PewDiePie (Felix Kjellberg). It is not a toy project or a celebrity side hustle: the codebase is real, runs on your own hardware, and bundles chat, autonomous agents, deep research, documents, notes, email, and model comparison into a single private dashboard. It competes with tools like LM Studio, Ollama, Open WebUI, and AnythingLLM, but covers more ground than any single one of them. Odysseus is a Chat GUI, an Agent Runner, and a Model Manager rolled into one. The philosophy is explicit: own your AI instead of renting it. The repository crossed over 79,000 GitHub stars on the dev branch within weeks of its May 2026 release, a signal that the approach resonates with developers and home-lab operators who want local control without sacrificing capability.

Odysseus gives you a browser-based interface that runs entirely on your machine. You connect it to local inference engines like Ollama, llama.cpp, LM Studio, or vLLM, or to cloud APIs from OpenAI, Anthropic, or OpenRouter. The core experience starts with chat: multi-turn conversations across any connected model, with persistent memory that the assistant builds and recalls across sessions. You can flip on agent mode to let the assistant plan, call tools, and work through multi-step tasks autonomously. Built-in tools include bash shell access, file operations, web search, and memory. You can also attach any MCP server for custom tool integration, toggling each tool per session.

Beyond chat, Odysseus includes a Deep Research module that gathers, reads, and synthesizes sources into a written report. The Compare view sends one prompt to several models at once and displays their answers side by side, blind or labeled. The Documents area is a writing-first editor with AI edits, suggestions, Markdown, HTML, CSV, and syntax highlighting. There is also an Email Assistant that connects over IMAP/SMTP for summaries, draft replies, auto-tagging, and spam triage. Notes, tasks, calendar, and a gallery/image editor round out the workspace. The app exposes no local API server by default; all interaction happens through the web UI, which is served from within the Docker container on http://localhost:7000.

Odysseus is free and open source under AGPL 3.0. No subscription, no telemetry. It runs on macOS, Windows, and Linux. The recommended install path is Docker: clone the repo, copy the environment file, and run docker compose up -d --build. Native install scripts exist for macOS and Windows (PowerShell) but are less tested. Confirmed GPU acceleration on Apple Silicon (MPS) and NVIDIA (CUDA). CPU-only works for smaller models. The app does not bundle an inference engine; you need Ollama, LM Studio, llama.cpp, or vLLM installed separately to serve local models. The Cookbook feature helps you pick and download models matched to your hardware, so you are not left guessing. There is no cloud dependency and no account required, unless you choose to connect external APIs.

All-in-one workspace. Chat, agents, research, documents, notes, email, calendar, and model comparison live in the same private dashboard. You do not need to juggle five apps.

Backend-agnostic. Point Odysseus at any local or cloud endpoint. It works with Ollama, llama.cpp, LM Studio, vLLM, OpenAI, Anthropic, OpenRouter, and any OpenAI-compatible API.

The Cookbook. Hardware-aware model recommendations across 270+ catalogued models. It shows which models fit your RAM and GPU, provides download links, and offers one-click serving configuration so you go from selection to running in fewer steps.

MCP support. Connect any Model Context Protocol server to extend the assistant with custom tools, databases, or APIs. Tools are toggled per conversation, giving you fine-grained control over what the agent can touch.

Agent mode with skills. The assistant can write, refine, and reuse its own skills. It becomes more capable over time without requiring you to handcraft prompts every session.

Deep Research. Multi-step research runs that gather sources, read them, and synthesize a report. All processing stays on your machine when using local models.

Privacy by default. No telemetry. No external calls unless you explicitly connect a cloud model. Data stays in your container. Third-party audits (e.g., Future Stack Reviews) confirmed that local-model chat turns produce no external network flows.

Private ChatGPT replacement. Run a full-featured AI workspace entirely on your own hardware with nothing leaving your machine. Best for people who need offline capability or have data they do not want to send to cloud providers.

Local model development and testing. Use the Compare view to test multiple models side by side before deciding which to deploy. The blind comparison mode removes bias.

Agentic research and document work. Run multi-step research and document editing with local models and tools. The agent can call web search, read files, and write output, all within the same UI.

Email triage and drafting. Connect an IMAP account for AI-powered summaries, drafts, and spam triage. Useful for power users who want to keep email processing local.

Who should look elsewhere. Non-technical users who want a five-minute installer will struggle with Docker and command-line setup. Anyone needing a sandboxed agent tool should be aware that the bash tool is not sandboxed and effectively grants shell access. Sensitive production deployments should audit the threat model.

1. Install prerequisites. Docker and Docker Compose. Optionally install Ollama or another local inference engine if you plan to run models locally.
2. Clone and configure.

1    git clone https://github.com/pewdiepie-archdaemon/odysseus.git
2    cd odysseus
3    cp .env.example .env

Edit .env to set a secure admin password and, optionally, your cloud API keys.

1. Launch.

1    docker compose up -d --build

The first admin password is printed in docker compose logs odysseus. Open http://localhost:7000.

1. Connect a model. Navigate to the Cookbook to download and serve a model matched to your hardware, or point Odysseus at an existing Ollama endpoint by entering its URL in Settings.
2. Start a conversation. The smallest useful step: open Chat, select a model, and send a message. Turn on agent mode or add MCP tools as you get comfortable.

The full setup guide, GPU notes, and HTTPS instructions are in docs/setup.md in the repo. The community is active on GitHub Issues and the dev branch receives daily updates.

Odysseus vs. LM Studio. LM Studio is a focused local model runner with a clean chat UI, but it does not include agents, research, email, or a document editor. Choose LM Studio if you only need to browse and chat with models. Choose Odysseus if you want a single app that also handles research, agents, notes, and email — and you are comfortable with Docker.

Odysseus vs. Open WebUI. Open WebUI is a popular self-hosted chat interface with RAG, tools, and multi-user support. It is lighter and easier to deploy. Odysseus offers more built-in features (email, calendar, image editor, skills system) but is heavier to run. If you need a straightforward chat and document Q&A tool, Open WebUI may be faster to set up. If you want a personal AI workspace that replaces several separate apps, Odysseus covers more ground.

Odysseus vs. Ollama. Ollama is an inference engine, not a full workspace. Odysseus needs Ollama (or another engine) to serve local models. They are complementary: Ollama handles the model serving, Odysseus provides the interface and tooling.

Pros. One private app covers chat, agents, research, documents, email, and more. Runs entirely on your hardware with no telemetry. Backend-agnostic, so it works with several local engines or cloud APIs. The Cookbook reduces the guesswork in model selection.

Cons. The breadth of features can feel overwhelming at first. Setup is Docker and command line, not a one-click installer. Getting the most out of it assumes capable local hardware. The agent bash tool has no sandbox — treat it as shell access. The dev branch can be unstable; use main for a more curated release.