Ingest Operation#
Think of it like a Kitchen#
You come home from the market with a bag of groceries (a source: article, transcript, conversation). You don't just dump the whole bag in the fridge — that's hoarding, not cooking.
Then you sort — vegetables here, proteins there. The LLM does this: reads the source, identifies 1-5 key concepts. That's like deciding "this bell pepper goes in the stir-fry, not the salad."
Then you prepare — wash, chop, season. The LLM writes a summary, picks tags, figures out [[wikilinks]] (cross-references) to existing dishes on your shelf.
Then you store — put each dish in its own labeled container. The ingest script creates or updates one .md file per concept. If the container already exists (you already have a "bell pepper" page), it merges the new info with the old — never duplicates.
Finally you update the menu — the script adds entries to _index.md so you can find everything later.
And you git commit — like taking a photo of your organized fridge. If you mess up tomorrow, you can always go back to this perfect state.
The Actual Flow#
Source text → Agent (LLM) extracts topics as JSON → ingest.py writes/updates wiki pages
→ auto-crosslinks via `[[Page Title]]`
→ updates _index.md
→ git commit
What Happens to Existing Pages (Merge Behavior)#
When the source talks about a concept that already has a wiki page, the script merges:
- Tags from old + new are combined (no duplicates)
- New aliases (alternate names) are added
- The body stays as-is (only new pages get the summary body)
- updated date refreshes
- The source field is preserved if it already has one
This means: re-reading the same article months later won't overwrite your added notes. Your refinements survive.
When to Ingest#
- You read an article that teaches you something
- You had a conversation that produced insight
- You solved a tricky bug and want to preserve the lesson
- You find a reference that connects two existing pages
Remember: Ingest is not archiving the source — it's digesting it. The raw text is disposable. The structured, cross-linked knowledge is what lasts.