Odysseus setup with Ollama, Gemma and hosted APIs

This walkthrough introduces Odysseus as a self-hosted AI workspace for chatting with models and using agents. The presenter demonstrates installation on an Apple Mac: clone the repository, enter its folder, run the setup script, and create an admin username and password. He also explains how binding to all interfaces makes the workspace accessible on the same network.

For the local LLM setup, the presenter uses the model menu's quick-start server scan to find Ollama. His machine runs Gemma 26B; he suggests smaller models for hardware that cannot run it. Chat mode handles conversation, while agent mode provides web search and shell access to execute commands on the computer. The presenter identifies OpenCode as the basis for the agents.

Cookbook recommends compatible models after scanning hardware and provides download and run controls. Other features include Tongyi Deep Research-based reports, blind comparison of two model responses, saved memories and skills. Notes can supply material to an agent, and the library collects items such as research reports. The presenter finds deep research useful but says it falls short of some proprietary options.

The final section shows provider settings for OpenRouter and NVIDIA NIM. The speaker also describes offline use with local models, though web search and hosted API routes depend on network access.