1 — Talk to the model
This runs on a GPU in this room. Not the cloud. Try fast, then try smart.
2 — What data is there?
Don't read a list. Ask. This is every data source in the room, including ones other teams made.
3 — Ask the documents
Policies, SOPs, manuals. Answers come back with citations.
Now make it yours
Open PROMPTS.md, pick a use case, and paste the prompt into your agent.
Everything above is ~40 lines of code in llm.js and kamiwaza.js. Go break it.