Skip to main content
A static Vite app. No backend, no database, no credentials required to run it.

​
Run

Opens on http://localhost:5173.

​
How it’s built

​
The seeded dataset

The demo data is generated, not recorded: 90 days, 24 people, 6 agents, roughly 5,850 records. It is deterministic, so every run of the demo tells the same story. Three beats are seeded and enforced by tests, which is what stops the narrative silently breaking as the generator evolves:
  1. A runaway loop — one agent, three days back, running at more than 3× its median
  2. A team roughly 40% over budget
  3. A right-sizing opportunity worth at least $2.5K per month
Anthropic model prices in the dataset are real list prices. Other vendors’ prices are plausible mocks. Keep that distinction in mind before quoting a cross-vendor comparison from a demo.

​
Recommendations are computed

Savings recommendations are derived at runtime from mock usage multiplied by list-price deltas — never hardcoded strings. The consequence that matters: change the underlying usage and the recommendations change with it, so the figure shown is always consistent with the data on screen.

​
Screenshots

Captures every view headlessly through the local Chrome install.

​
Security posture

The prototype ships no backend and no credentials. The optional Anthropic connector reads an admin key from the environment at script runtime and writes its snapshot into public/live/, which is gitignored. The key is never persisted. The CSV import parses in-browser and uploads nothing.

Connector setup

How to wire up the Anthropic connector and the ChatGPT Enterprise CSV import.