Proof, dated

VidMee fleet stats

Live, dated telemetry from VidMee running on its own dogfood and the Hermes QA controller, not customer usage. Every number is a count or median over our real production traffic; the machine-readable source is GET api.vidmee.ai/v1/stats, re-counted here on load. Counts grow.

Loading live figures…

Headline

…tool calls runon our fleet, all-time

…median tool callcompleted, end-to-end

…vision analysesjudgments produced

axe-core WCAG scans

Tool calls by tool

Tool	Calls
loading…

Generation

…provider generationsimage + video attempts

…median cost / generationprovider-billed USD

…median tokens / vision callcaptured from provider usage

gemini-flashvision modelrouted, with fallback

Token-per-call medians populate as new vision calls run after the 2026-07-12 instrumentation; the field reads null until the first measured window closes.

Latency snapshot (measured 2026-07-12)

9.9 smedian tool callend-to-end

2.2 smedian page renderrender pipeline

5.7 smedian vision analysisjudgment

43tools, one MCPas of 2026-07-11

Changelog

2026-07-12Instrumented per-call vision token + cost telemetry; published this live stats page (/v1/stats).
2026-07-11Tool count 39 → 43 (added storyboard_video, list_providers, list_models, and namespaced discovery tools).
2026-07-04Shipped axe-core WCAG accessibility scans in the render pipeline.
2026-06-28First production tool calls recorded.