Perplexity Model Watcher by MaxScorpy

Perplexity routes your question to different models and doesn't always
make it obvious which one actually wrote the answer — and sometimes it
isn't the model you selected. Perplexity Model Watcher reads the real
answer and shows the model that produced it, right on the response.

WHAT IT DOES

• A small badge on each answer names the model that actually replied,
  with human-readable names instead of raw internal identifiers.
• Mismatch alert: when the model that answered differs from the one you
  requested, the badge flags it — so you know when Perplexity quietly
  fell back to a different model.
• Draggable badge: move it wherever it doesn't get in your way; its
  position is remembered between sessions. Minimize it to a compact pill
  when you just want a glance.
• Survives Perplexity's single-page navigation and re-renders — the
  badge re-attaches itself as you move between threads and follow-ups.

WHY IT'S PRIVATE

• Runs only on perplexity.ai. It reads the answer stream on the page to
  identify the model — nothing is sent anywhere.
• Zero telemetry, zero analytics, zero remote calls.
• Only the "storage" permission, used to remember the badge position and
  your preferences locally.
• AMO data_collection_permissions disclosure is set to "none".

CREDITS

Built on Model Watcher by the Model Watcher contributors
(https://github.com/apix7/perplexity-model-watcher), used under the MIT
License. The licence and an attribution notice ship inside the add-on
(LICENSE and NOTICE.md). What this version changes: the answer stream is read
progressively and finalised on a quiet timeout, so the badge cannot stay stuck
on "analyzing" after a follow-up, and an explicit idle/active/done phase drives
it. The draggable, minimisable badge and its remembered position come from the
original project.

LANGUAGES

Interface in Russian; model names are shown as-is (GPT, Claude, Sonar,
Grok, etc.).