A note is not useful because I saved it.

It is useful if I can find it when it changes what I do next.

That is the whole problem with connecting Obsidian to an agent. I do not need more places to dump thoughts. I need the right context to show up without trusting my memory or letting an agent guess.

The vault holds the mess

Obsidian is good at messy human material.

Drafts. Research notes. Half-built theories. Meeting fragments. Trade lessons. Source links. Questions I do not know how to answer yet.

That mess is useful, but it is not automatically operational.

A warehouse full of parts is still just a warehouse. It becomes useful when the shelves are labeled and somebody knows where to look.

Memory needs a chain of command

Bad agent memory is a junk drawer with a confident voice.

It remembers too much, too vaguely, and just enough to sound sure of itself when it should be checking the file.

The better setup is simpler:

memory for stable facts
notes for context
files for authority
tools for verification
receipts for continuity

That order matters.

If the vault says one thing and the repo says another, the repo probably wins. If old memory disagrees with the live system, the live system wins. If nobody wrote down the decision, maybe it was not a decision yet.

Search is not recall

A search result is a doorway, not an answer.

The agent still has to read around the match, understand why the note exists, and separate old task residue from context that still matters.

There is a big difference between “find every mention of this ticker” and “tell me what thesis I was testing.”

The second one needs better notes.

Make the next session cheaper

That is the point.

A good note should make the next build easier. A good receipt should keep the next session from repeating discovery. A good instruction should keep the same mistake from happening twice.

The vault should become working memory, not a sentimental attic.

Not AI as oracle.

AI as a librarian that knows when to open the cabinet before making claims.