odysseus  by pewdiepie-archdaemon

Odysseus provides a self-hosted, local-first AI workspace, replicating ChatGPT/Claude UX. It targets users prioritizing privacy and control, offering a feature-rich, open-source alternative running entirely on user hardware.

How It Works

FastAPI-based, Odysseus integrates chat with local models (vLLM, llama.cpp, Ollama) and APIs, plus an agent system using shell, file, and web tools via MCP. Its "Cookbook" handles VRAM-aware model serving (GGUF/FP8/AWQ). Modules include multi-step "Deep Research" synthesis, blind model comparison, AI-assisted document editing, persistent memory (ChromaDB), AI email triage, and CalDAV calendar sync.

Quick Start & Requirements

Requires Python 3.11+. Docker recommended.

• Docker: Clone repo, cp .env.example .env, docker compose up -d --build. Access http://localhost:7000.
• Native Linux/macOS: Clone, setup Python venv, pip install -r requirements.txt, python setup.py, python -m uvicorn app:app --host 0.0.0.0 --port 7000. Access http://127.0.0.1:7000.
• Native macOS (Apple Silicon): Use ./start-macos.sh for Metal GPU support.
• Native Windows: Use .\launch-windows.ps1 or manual Python setup.
• Links: Demo tour: docs/index.html. Contributing: CONTRIBUTING.md. Roadmap: ROADMAP.md.

Highlighted Details

• Cookbook Model Serving: Serves GGUF, FP8, AWQ models with VRAM awareness (vLLM/llama.cpp).
• Agent & MCP Integration: Agents use tools, including browser automation (Playwright/MCP).
• Productivity Suite: AI email triage, notes, tasks, CalDAV calendar sync.
• Mobile-First Design: Responsive UI, PWA support.
• Local-First Data: All user data stored locally, Git-ignored.

Maintenance & Community

Provides CONTRIBUTING.md and ROADMAP.md for development guidance and future plans. Specific community channels (e.g., Discord, Slack) or notable maintainer/sponsor information are not detailed in the provided README.

Licensing & Compatibility

Licensed under the MIT license, generally permissive for commercial use. Security notes advise HTTPS and reverse proxy for network deployments to protect credentials and API tokens.

Limitations & Caveats

• macOS GPU: Docker on macOS lacks Metal GPU support; native execution is required. vLLM/SGLang are CUDA/ROCm-only and do not run on macOS. MLX models are unsupported.
• Security Configuration: Requires careful setup (HTTPS, reverse proxy, authentication, user privileges) for deployments beyond localhost to prevent exposure of sensitive data.
• MCP Dependencies: Browser MCP functionality requires an initial npx command to install Playwright.
• "Jank and Fun": The project's self-description suggests a less polished, more experimental feel compared to commercial offerings.