Show & Tell
In BuildLab project

UnderDock / NICOL

Sovereign, offline-first AI orchestration on our own hardware. NICOL leads the room; a person approves every change.

UnderDock / NICOL logo

About

UnderDock runs AI orchestration across our own fleet: a workstation and two NVIDIA DGX Sparks, with no dependency on a cloud endpoint for the work that matters. NICOL is the lead agent. She runs the roundtable with UnderdogLM, the developer seat and Penny, the cost tracker, gathers the evidence, and turns findings into proposed changes. Nothing ships on its own. Every change arrives as a card with its basis attached, and a person approves it or sends it back. In the demo, NICOL pulls 14 days of real usage records, Penny builds the report, and the finding becomes a proposed fix waiting on the operator's sign-off. Designed for sites where the network is not the answer and the data does not leave the room.

Screens

What it looks like

What’s new

Updates

  1. NICOL now reviews its own diffs before asking

    The build agent reads its proposed change set, flags the lines it is least sure about, and hands that context to the reviewer — so the approval step is a decision, not a rubber stamp.

    NICOL now reviews its own diffs before asking media