PennyOS app

Chat

A conversation with a local model on your server that finds the closest sections of your wiki and answers from them, naming the pages it used.

• Home key: q
• Needs: Needs your server
• Runs on: Your server, with Ollama
• Offline: Cannot answer

What it does

Chat is the private, local counterpart to Ask. Your question never leaves your server. A local model reads the parts of your wiki that best match the question and answers from them, then ends with a Pages: line so you know where the answer came from.

Keys

How it works

1. Each turn goes through the same SSH gate as Ask, to the localask shape.
2. On the server, a retrieval script embeds the question, finds the closest wiki sections, and passes them to the local model with your last six turns.
3. It streams back which pages it is reading, then the answer, then the pages it used. The footer shows the page names while it works.

It reads the vault and never writes to it.

Direct mode

Set chat_direct = true to skip the vault and talk to Ollama over plain HTTP at ollama_url with ollama_model. That is conversation with no context, and it is off by default for that reason. It exists for a device that can reach Ollama on your tailnet but not the gate.

Needs on the server

• Ollama with a model pulled. Bind it to a private address such as your Tailscale IP, never to every interface.
• A retrieval script at tools/local_ask.py in your vault, which takes a question and streams back the pages it read and the answer. PennyOS does not include one; see Working with Obsidian. Without it, Chat reports the error from the server.