Notes from the building.
Plain accounts of how the system is configured, where its answers come from, and how an answer gets checked before anyone acts on it.
All posts
- What “local AI” actually means: connected, air-gapped, and in between · 6 min read"Local," "on-prem" and "private" can mean different things: where the model runs, where the index lives, and what crosses your network in connected, isolated and in-between setups.
- How company data ends up in AI tools · 6 min readStaff pasting work, sharing features, provider bugs, vendor breaches and court orders: the ordinary routes company information takes into AI tools, and what an organization controls.
- Shadow AI: when staff paste work into public AI tools · 5 min readWhy staff paste work into AI tools nobody approved, what has gone wrong when they did, and the practical steps organizations take to close the gap.
- What we will not claim · 4 min readNo accuracy percentages, availability figures or cost-savings promises - what we'll say instead, and what a consultation is actually for.
- A physical address for intelligence · 6 min readWhy the system is a box on your floor and not a tenant in someone else's cloud, and what that distinction actually buys you.
- The question, the source, the review · 5 min readA plain description of how a query becomes an answer: the question, the approved source, the qualified output, and the human review that follows.
- Configured around the work · 5 min readA system configured around a specific workflow can be checked in a way general-purpose infrastructure asked to do everything at once cannot.
- Intelligence, in place · 5 min readWhat changes when the hardware that runs your AI sits on your floor instead of in someone else's data center: day-to-day answers stay on site.
Put local AI to work
where it matters most.
Bring the question you would ask your own documents.