Capturing Tasks by Voice

Next to the Add a task… box there's a microphone button. Press it, say what you need to do, press stop — and KnotDo fills in the task for you: title, due date, priority, and any extra detail as notes. Nothing is created until you press Add. Voice capture drafts the task; you get the final say. If a word was misheard, fix it in the box before saving. Two separate things happen behind the scenes:

• Speech to text — your recording becomes a transcript.
• Text to task — the transcript becomes a structured task.

The inline microphone is small on a phone, so mobile gets a dedicated + button in the bottom-right corner — within thumb reach on any screen. Tap it and pick:

• Type it — opens the task box with the keyboard up
• Say it — starts recording straight away

The + button follows you across the app. If you tap it somewhere without a task box — a Kanban board, Calendar, or Settings — it takes you to Today and picks up where you left off rather than doing nothing. Tap the backdrop, the ×, or press Escape to dismiss the sheet without adding anything.

AI transcription costs money to run, so KnotDo doesn't bundle it — you connect your own provider account and the usage bills to you directly, at whatever that provider charges. No markup, no per-seat AI fee, and you can disconnect at any time. Go to Settings → AI Keys and paste a key from any of:

• Claude (Anthropic) — console.anthropic.com
• ChatGPT (OpenAI) — platform.openai.com
• Gemini (Google) — aistudio.google.com

Keys are encrypted before they're stored and are never sent back to your browser — the page only shows which providers are connected.

Claude has no speech-to-text API, so it can't transcribe audio. It's an excellent choice for turning the transcript into a task, but it needs a transcript to work from.

• OpenAI or Gemini connected — your recording is transcribed on the server. This is the most accurate option and works in any browser.
• Only Claude connected — transcription falls back to your browser's built-in speech recognition (Chrome, Edge, or Safari), then Claude turns it into a task.
• Nothing connected — the browser still transcribes, and the raw transcript becomes the task title with no parsing.

Once you've connected more than one provider, pick which one writes the task using the buttons at the bottom of the AI Keys tab.

Speak it the way you'd say it to a person: "Call the dentist tomorrow about the crown, this is urgent." That becomes a task titled "Call the dentist", due tomorrow, priority Urgent, with "Ask about the crown" in the notes. Relative dates are resolved against your calendar day, not UTC — so "tomorrow" means tomorrow where you are. Recordings stop automatically at five minutes. The audio itself is never stored: it's sent for transcription and discarded.

No microphone button. Your browser doesn't support audio recording, or the page isn't on HTTPS. "Microphone blocked" — and no permission prompt appeared. That combination means your browser already has a stored decision for knotdo.app, so it rejects instantly instead of asking again. Dismissing the prompt a couple of times counts as a block. Click the icon at the left of the address bar → Site settings → Microphone → Allow, then reload. If it's already allowed there, the block is above the browser: on Windows check Settings → Privacy & security → Microphone and turn on both "Microphone access" and "Let desktop apps access your microphone". If you installed KnotDo as an app, the standalone window keeps its own permission state — grant it there too. "No microphone found" / "in use by another app". These are not permission problems: plug in or select an input device in your system sound settings, or close whatever is holding the mic (Teams, Zoom, Discord). "That API key was rejected." The key is wrong, revoked, or out of credit. Check it in your provider's console and paste a fresh one into Settings → AI Keys. Nothing was heard. With no OpenAI or Gemini key connected, transcription relies on the browser — which needs Chrome, Edge, or Safari. Firefox has no built-in speech recognition, so connect a transcription key to use voice capture there. "Too many recordings in a row." Voice capture is limited to 20 recordings every five minutes to protect your API credit from a runaway loop. Wait a few minutes.