EmberType logo - voice-to-text dictation software for macOS

AI Enhancement & Configuring Models

Pass your transcription through an AI model to clean it up, reformat it, or turn spoken commands into written responses.

What Is AI Enhancement?

AI Enhancement is an optional post-processing step that runs your Whisper transcript through a Large Language Model (LLM). Whisper turns your voice into raw text. AI Enhancement then reshapes that text — fixing filler words, adding punctuation, reformatting as a list, or even responding to a spoken instruction — based on the prompt you select.

Think of it as a two-stage pipeline:

Voice → Whisper transcription → AI Enhancement (optional) → Final text

Enhancement is disabled by default. When it's off, you get Whisper's raw output. When it's on, your raw transcript is sent to whichever AI provider you've configured, processed with the active prompt, and the result is what gets typed into your app.

Enabling AI Enhancement

  1. Open the AI Enhancement panel
    Click the sparkle icon in the mini recorder during dictation, or open EmberType and navigate to the AI Enhancement section.
  2. Toggle "Enable Enhancement" on
    Once enabled, the prompt grid and provider options become active.
  3. Configure an AI provider
    Enter an API key for at least one provider (see below) — or set up a local model through Ollama for fully offline enhancement.
  4. Pick a mode
    Select Default, Assistant, or a custom prompt. The selected prompt becomes the active mode until you change it.

Built-In Modes: Default vs. Assistant

EmberType ships with two predefined modes. They behave very differently, and picking the right one matters.

Default

Clean up what I said

A transcription enhancer. It polishes your raw transcript — fixes filler words, run-ons, punctuation, and speech recognition errors — while preserving your meaning and tone. It never responds to what you said, even if you spoke a question out loud.

This is the right mode for dictating emails, messages, documents, notes, or anywhere you just want a cleaner version of your own words.

You say "um so I was thinking like maybe we should uh push the meeting to Thursday because you know the client's gonna be out" You get "I was thinking maybe we should push the meeting to Thursday because the client is going to be out."

Assistant

Do what I said

A true AI assistant. It treats what you spoke as a request and generates a response. Ask a question, get an answer. Ask for a draft, get a draft. Ask for a summary, get a summary. The LLM responds directly with no preamble or sign-off.

This is the right mode for on-the-fly AI tasks driven by voice — drafting, summarizing, translating, explaining, rewriting.

You say "Write a two-sentence reply to my boss letting her know I'll be out tomorrow for a doctor's appointment." You get "Hi, I wanted to let you know I'll be out tomorrow for a doctor's appointment. I'll follow up on any urgent items when I'm back."

Quick comparison

Behavior Default Assistant
Cleans filler words Yes Not applicable — it generates fresh text
Answers questions you ask No (treats them as text to clean) Yes
Generates new content No Yes
Preserves your exact wording Mostly — only light edits No — produces a response, not a transcript
Best for Dictation, messages, docs Drafting, summarizing, Q&A

The shortcut: Default = clean up what I said. Assistant = do what I said.

Custom Prompts

Default and Assistant are the two built-in modes, but you can create as many custom prompts as you want. A custom prompt is just a set of instructions the LLM follows when processing your transcript.

Creating a custom prompt

  1. Open the AI Enhancement panel
  2. Click the + button in the prompt grid
  3. Give it a title, description, and icon
  4. Write your prompt instructions
  5. Save — it will appear in your prompt grid alongside Default and Assistant

Good custom prompt ideas

You can switch between prompts at any time by clicking a different card in the prompt grid. The selected prompt becomes your active mode.

Configuring AI Providers

AI Enhancement needs an LLM to do the actual processing. EmberType supports several providers — pick whichever fits your workflow, budget, and privacy preferences.

Cloud providers (API key required)

Local (fully offline)

Adding an API key

  1. Open the AI Enhancement panel in EmberType
  2. Find the API Key Management section
  3. Paste your API key into the field for the provider you want to use
  4. Select a model from the dropdown that appears

Privacy: API keys are stored in your macOS Keychain, never in plaintext. When you use a cloud provider, your transcript is sent to that provider for processing. If you need full privacy, use Ollama for local enhancement — nothing leaves your Mac.

Clipboard and Screen Context

AI Enhancement can be given additional context to improve accuracy and relevance:

Both toggles appear under the main Enable Enhancement switch and can be turned on or off independently. They work with any mode (Default, Assistant, or custom).

Privacy and Data Handling

EmberType itself is 100% offline — Whisper transcription always runs locally on your Mac. AI Enhancement is the one feature that can send data to a third party, and only if you choose a cloud provider.

Read the full Privacy Policy for more details.

Troubleshooting

Enhancement is on but nothing is happening

Assistant mode isn't responding, it's just cleaning up my text

You're probably on Default mode. Open the AI Enhancement panel and click the Assistant card to make it the active mode. The selected card is highlighted.

Default mode is trying to answer my questions instead of cleaning them up

You're probably on Assistant mode. Switch to Default by clicking the Default card in the prompt grid.

The enhanced output lost important words or numbers

Enhancement is too slow

I want different behavior than Default or Assistant offer

Create a custom prompt. See the Custom Prompts section above.

Back to Documentation