Support & FAQ

This is almost always a permissions issue. Spill It! needs two macOS permissions to work:

1. Microphone — so it can hear you
2. Accessibility — so it can type at your cursor and read screen context

Go to System Settings → Privacy & Security and check both are enabled for Spill It!.

When you reinstall, update, or move the app to a different folder, macOS invalidates the old permission because the app binary has changed.

To fix this:

1. Open System Settings → Privacy & Security → Accessibility
2. Find Spill It! in the list
3. Toggle it OFF
4. Toggle it ON again

This re-authorizes the current binary. You only need to do this once after each reinstall or update.

For the first launch only — to download the speech model (~300 MB). After that, Spill It! works fully offline.

If you use AI polish, that does require internet since it calls your configured API provider. But the core transcription never needs a connection.

Bluetooth microphones often use a low-quality codec (HFP/SCO at 8kHz) when the mic is active, which significantly reduces transcription accuracy.

For the best results, use your Mac's built-in microphone or a wired mic. If you need to use Bluetooth, check that your headset is in "high quality" or "music" mode — some headsets let you choose.

This used to happen when the app tried to transcribe silence or background noise. We've fixed this — the app now uses voice activity detection to discard audio with no speech in it.

If you still see this, make sure you're on the latest version (check for updates in the app menu). If the issue persists, it may be that your microphone is picking up very faint sounds — try in a quieter environment or adjust your mic sensitivity.

Yes. Spill It! uses automatic gain control that boosts quiet signals. Whispering works at normal desk distance — you don't need to be close to the mic. The people around you won't hear you dictating.

Spill It! reads what's visible on your screen — names, file paths, identifiers — and uses them to correct the transcription. If a name or term is on screen (in a chat, email, or IDE), it should be picked up automatically.

If it's not on screen, the speech model transcribes what it hears phonetically. For unusual names or niche terms, having them visible somewhere on screen is the best way to get accurate results.

Number conversion works for most patterns: "forty two" → 42, "three thirty PM" → 3:30 PM, "March thirty first" → March 31st.

Some patterns are intentionally left as words to avoid false conversions. For example, "three twenty five" might look like a time (3:25) so it's left as words rather than guessing wrong.

Currencies require you to say the currency word: "fifty dollars" → $50. If you just say "fifty" without "dollars", it stays as the number 50.

English only at this time. The speech model, text processing, and all formatting rules are built for English. Multi-language support is something we'd like to add, but it's not available yet.

Local transcription (always on) runs entirely on your Mac. It handles filler removal, stutter detection, number formatting, path assembly, and context-aware output. It's fast and private — nothing leaves your machine.

AI polish (optional) sends your transcribed text to an external AI provider you choose (OpenAI, Anthropic, Cerebras, etc.) for additional refinement — paragraph structure, question marks, more aggressive cleanup. You bring your own API key. We never see the text.

AI polish adds latency (typically 1–3 seconds depending on text length and provider) and costs a small amount per use (roughly $0.001–$0.01 per transcription depending on the model).

Your voice is always processed on your Mac. No audio ever leaves your machine — not to us, not to anyone.

If you enable AI polish, the transcribed text (not audio) is sent to the API provider you configure. It goes directly from your Mac to their servers — our servers never see it. What the provider does with that text depends on their data retention policy.

We recommend checking your provider's settings and opting out of data training if that option is available.

Very little:

• License verification — a call to check your license status. Contains your license key, not your transcriptions.
• Crash reports (if you enable them) — sent to Sentry. Contains the crash stack trace and your macOS/app version. Never includes audio, text, or anything you've said.
• Analytics (off by default, opt-in) — sent to PostHog. Contains feature usage events like "opened settings". Never includes transcription content.

That's it. No audio, no transcriptions, no prompts, no clipboard contents.

After purchasing through Paddle, the license should activate automatically. If it doesn't:

1. Check your email for the Paddle receipt — it contains your activation key
2. In Spill It!, go to Settings → General and paste the activation key
3. If the app was closed during purchase, relaunch it — it checks for pending activations on startup

If none of that works, email us at hello@tryspillit.com with your Paddle order number and we'll sort it out.

AI polish requires your own API key from a supported provider (OpenAI, Anthropic, Cerebras, etc). Go to Settings → Cloud AI and make sure your API key is entered and the feature is enabled.

If you just purchased and the app doesn't show your paid tier, try relaunching — it checks for pending activations on startup.

Your license is tied to your Paddle order, not to a specific machine. You can install and use it on your personal Macs. If you run into activation issues on a second machine, contact us.

macOS 14 (Sonoma) and later, on Apple Silicon Macs. Intel Macs are not supported — the on-device speech model requires Apple's GPU for real-time performance.

Spill It! checks for updates automatically using Sparkle. You'll see a notification when a new version is available. You can also check manually from the app's menu bar icon → Check for Updates.

After updating, you may need to re-toggle Accessibility permissions (see the first question above).

A few things to check:

• First use after idle: If the app has been idle for a few minutes, the first transcription may take slightly longer as the model warms up. Subsequent ones will be fast.
• AI polish enabled: If AI polish is on, there's additional latency (1–3 seconds) while the text is refined by your API provider. This is normal — it's a network round-trip.
• macOS App Nap: In rare cases, macOS may put the app to sleep if it thinks it's inactive. We prevent this, but if you notice sluggishness, try keeping the app in the foreground briefly.