Plumadex

Most bird identification apps send audio to a remote server, which requires network connectivity, introduces latency, and means your recordings leave your device. Plumadex runs inference locally — the model lives in the browser, the audio never leaves.

Press Record. The app captures ambient sound, runs it through an on-device TensorFlow Lite model, and returns species matches in under a second. No account required, no network call for the inference step. Works in the field where signal is weak or absent.

6521 species across 7 languages. Common names, taxonomic data, and species profiles are localised into English, Russian, German, French, Spanish, Chinese, and Japanese — making the app usable for birders and researchers across Europe, Asia, and the Americas.

The model is trained on spectrograms derived from field recordings. It handles overlapping calls, background noise, and partial vocalizations without needing clean isolated audio.

The app is a PWA — installable from the browser with no app store friction, works offline after the first load, and receives updates without user action. An account system manages species lists and observation logs for users who want to track sightings across sessions.

Plumadex is live and free to use.