Building HoneyDrunk.Lore: My LLM Wiki and Daily News Blast
I read too much. I follow model updates, agent news, architecture posts, and security research. Most of this information does not matter right now. It matters later.
Normal bookmarks fail me. They save links but lose the meaning. Chat history saves the talk but loses the structure. RAG gives me chunks but makes me rebuild my thoughts every time.
I built HoneyDrunk.Lore to solve this. It is an LLM wiki for my studio. It turns raw information into a compiled knowledge system.
The system uses a simple pipeline:
- Raw sources land in an evidence locker.
- Agents read them and extract claims.
- The wiki updates topic pages and links concepts.
- A maintenance loop lints the data to find contradictions or gaps.
This is not just a pile of summaries. It is a maintained artifact.
The system also produces a daily news blast for Discord. It picks the top 10 web stories and top 10 social posts. Each item includes:
- A short summary.
- The original URL.
- A specific angle on why it matters to my work.
I treat social media as early signal only. A tweet might report a launch first. But the wiki waits for an official blog post or documentation before it treats that info as a durable fact. This separates "I saw a thing" from "the wiki knows a thing."
Lore is not agent memory. It is not governance. It is source-backed decision support. If the wiki makes a claim, it must point to the source and show its confidence level.
Search asks if you can find a thing. Lore asks if the thing has been digested into what you already know. One retrieves. The other compounds.
I am building this for the long term. I want a system that stays warm between sessions and grows as I learn.
Source: https://dev.to/tatted_dev/building-honeydrunklore-my-llm-wiki-and-daily-news-blast-2pfl
Optional learning community: https://t.me/GyaanSetuAi