The first version copied a research workflow too literally.
I tried to copy a public research workflow wholesale. It worked in theory, but in practice it was too token-heavy, too loose, and too dependent on large source dumps.
Knowledge System
Wiki LLM
A knowledge pipeline that turns raw source material into durable context, topic bibles, summaries, and reusable workflow assets.
Raw notes do not compound by themselves.
Separate vaults all need the same thing: a way to turn source material into reusable context instead of isolated storage.
Visual Explanation
How the wiki becomes working context.
The wiki starts with raw sources and project notes. The walkthrough shows how those inputs become concept pages, topic bibles, build guidance, media research packets, linted knowledge, and daily digests.
raw inbox
normalized source pages
claim routing
topic bibles
nightly maintenance
Inputs
Processing
Knowledge Base
Reusable Outputs
Maintenance
The folder system improved, but the workflow stayed manual.
The second version introduced a more structured vault layout with `00`, `10`, `20`, and later resource layers. That made the knowledge base easier to reason about, but I was still copying article text and transcripts into the vault by hand and asking OpenRouter to analyze them.
Python scripts reduced friction, but ingestion still depended on me.
The third version added Python scripts for normalization, claim routing, page updates, and topic-bible assembly. It was materially better, but still local and operator-driven: I had to sit in front of the computer, drop files into raw folders, and trigger ingestion myself.
The harness turned the wiki into an actual pipeline.
The fourth version wrapped the workflow in a harness. Now I can drop links into Discord and let the system handle capture, extraction, sorting, page merges, deduping, linting, and daily digest delivery without babysitting the ingestion loop.
The wiki stopped being a note archive and became infrastructure.
A topic bible can now feed directly into other builds, whether that means guiding the local harness, generating project checklists, or packaging a clean context layer for another tool.