Step-by-step guides

Update the control root first, then every other machine. What carries over, and the few things to do after.

Use Gemini, OpenAI, OpenRouter, ElevenLabs, or your Claude Code or Codex subscription beside your local models, through one address.

Let models search the web through Brave, Google Search or SearXNG, and choose the order Eugene tries them.

Serve embeddings and reranking from your own machine, for Workbench and for Open WebUI's document search.

Check which models fit this machine's graphics memory (VRAM) before you download, add a model hub with its own token, and delete a model safely.

Point Claude Code or Codex at your own local models, with one address and a key from the console.

Answer from your own documents with local models, in Workbench, Open WebUI or your own code.

Install Eugene's chat app on any of your machines, and sign in with Eugene.

Keep notes Workbench gives the model in your chats, see what each chat sent, and choose what models outside your machines get.

Attach a file to one chat, or keep documents and folders to search, with each passage cited by file and page.

Grant a folder, add MCP tool servers, and approve each call in the chat.

Let Workbench work on a person's own machine, wherever it is, with no incoming connections.

Keep a second machine ready to take over as the control root.

Optional: put the console and Workbench behind one HTTPS port, each with its own name.