name: Browser speech description: Free dictation using your browser's speech recognition. Audio may go to the browser vendor. version: 1.0.0 type: extension entry: plugin.py plugin_api: "2.0" runtime: python3 os: any

net: the browser sends the owner's audio to its vendor. It is not on by default (skills/builtin.json).

permissions: [page_module, net] when_to_use: The owner wants to dictate a draft using browser speech recognition. model_experience: what_model_sees: Only the draft text the owner chooses to send; no new tools. token_effect: No model tokens for recognition or while idle. contributes:

• id: speech

provides: stt label: Browser speech title: "Free. Your browser recognizes speech: Chrome sends audio to Google, Edge to Microsoft, Safari to Apple (on device when available)." runtime: client destination: vendor:Browser speech service render: {kind: host, entry: ui/stt.js}

Browser speech

Find this engine in the composer's “+” → “Add dictation”: it is not installed by default, and no audio leaves the page until it is chosen there as the dictation engine. Dictation writes an editable draft; only Send submits it.

Chrome recognition sends audio to Google, Edge to Microsoft. Safari uses Apple's speech service, on device when available. Other engines may use an online service; the browser controls processing, availability and language support. This skill makes no promise of local recognition. No model is called and no model tokens are used. A browser without speech recognition offers another engine instead.