A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

Here's the full setup to make Pi use OpenRouter by default, with your API key stored locally inside Pi's own config (no system-wide env var needed):

Step 1 — Create Pi's config directory (if it doesn't exist)

mkdir -p ~/.pi/agent

📺 Видео-версия гайда

1. Подключаемся к серверу:

I wanted to listen to The Bible using amazing Dwell app which sadly only has an Android or iOS app currently. 👎 So for this we need to use one of the cancerware CCP owned Chinese 🇨🇳 Android emulators. Yay! /s 🙄

Let's at least decrapify this thing as much as we can yeah? Optional/situational lines are in italics but these aren't required and may not work.

These below steps should work similarly for Nox/MEmu etc. because all these emulators are all owned by the same crapware company (see bottom of this gist), but you can older find guides for Nox here and MEMu here.

how to leverage oracle's temping offers

The limits of the free tier say that you can create up to 4 instances.

• x2 x86 instances (2core/1g)
• x2 ampere instances (with 4core/24g spread between them)
• 200GB total boot volume space across all intances (minimum of 50G per instance)