60-second walkthrough

The problem: prompts break silently in production

Most teams store prompts in Notion, spreadsheets, or .env files. When behaviour changes, nobody knows which edit caused it — or whether OpenAI updated the model.

PromptVault fixes three things in one workflow:

  1. Version — every edit is tracked with diffs
  2. Check — structural tests run before you promote
  3. Monitor — alerts when OpenAI model metadata changes

Sample data below — no account required. Sign up to run this on your own prompts.

1 — Version history

v4Add empathy instruction for premium tier
gpt-4oproduction4/4
v3Shorten response to 3 paragraphs
gpt-4o6/8 ↓
v2Initial support tone
gpt-4ostaging8/8

2 — Character diff (v2 → v3)

v2

Respond in no more than five paragraphs.

v3

Respond in exactly three paragraphs.

3 — Structural test results

Response mentions paragraph limitcontains · pass
Opening phrase exact matchexact_match · fail
Empathy instruction presentcontains · pass

Contains and exact_match validate rendered prompt text — not model output. LLM-as-judge is available for manual output grading.

4 — Model Watch event log

Watching gpt-4o · threshold 80%

Model metadata updated

snapshot: 1737201600 → 1740000000

2d ago

Structural tests triggered

pass rate 87% — above threshold

2d ago

Alert sent

model deprecated — gpt-3.5-turbo-0613 removed

5d ago

Model Watch monitors provider metadata and availability — not silent behavioural changes behind an unchanged model alias.

Run this on your own prompts

Free tier: 3 prompts, structural checks, and Model Watch for OpenAI models.

Alerts use Resend · polling runs daily on Vercel Cron