WHERE THIS TEST ENDED
This update explains the intended knowledge architecture. It does not demonstrate a complete question-to-answer trip through the whole system.
- Recorded information
- Transcripts and notes
- Knowledge on the mini
- Context for Fletcher
Start with the information I actually need
I use Obsidian as the place for my notes and describe a recording-to-transcript process that makes spoken information available to my agents. The point is to give Fletcher relevant context about my work, rather than expect a small computer to know everything by itself.
The video shows a graph of that knowledge and describes skills that keep it updated across the AI tools I use. This is a description of the knowledge workflow, not a benchmark of its accuracy.
Give the machines different roles
The Pi is the device at the desk. In the setup I describe, it handles the local assistant experience and can use a local model. The Mac mini runs Hermes and holds the agent side of the knowledge workflow. The Echo is the device I speak into.
When a request needs context from my second brain, the intended route reaches the knowledge system on the mini. That is different from asking the Pi to run every model, hold every note, and do the hand tracking alone.
What would make this useful
My example is asking about something relevant to my own schedule or project. The answer needs the right source information, not just fluent wording.
To prove this complete route, a later test needs to start with a real spoken question, retrieve the matching knowledge, and produce the correct reply through the desk. This video explains why I am connecting those pieces; it does not show that full test.
About this breakdown
Matched to this original TikTok, including its spoken captions. The result above describes this episode; later work is identified separately.
Breakdown checked September 6, 2026.