Action Mode

Most of the time dictation types what you say, word for word. Action Mode is the exception. Start an utterance with the wake word and Hyprcore treats the rest as an instruction—it hands your words to an LLM, runs them, and pastes the finished result instead of the transcript.

Say "Hyper, write a friendly reply saying I'll be five minutes late" and a polished reply lands where your cursor is. No app switch, no copy-paste.

How it works

1. You dictate as normal, but lead with the wake word: "Hyper, …".

2. Hyprcore transcribes the audio, then checks whether the transcript begins with a wake word.

3. If it does, the wake word is stripped and everything after it becomes the instruction.

4. The instruction runs through your post-processing LLM (the same provider you set for post-processing—Claude, GPT, Llama, your own Ollama, and so on).

5. The result is pasted where your cursor is. The raw transcript is not pasted.

If any text is selected when you speak, Hyprcore passes it along so the instruction can act on it—say "Hyper, make this more concise" with a paragraph highlighted and it rewrites that paragraph in place.

The wake word

The built-in wake word is Hyper. Because speech-to-text often mangles it, Hyprcore also matches the common phonetic variants hiper and high per—you don't have to configure these, they always work.

A few rules keep it from firing by accident:

• It has to lead the utterance. The wake word must appear within the first three words, so a natural greeting or filler ("Hey", "OK so", "Good morning") is skipped—"Hey Hyper, draft a thank-you note" works. Say "hyper" deeper into a sentence and it's treated as ordinary dictation.

• It has to be a whole word. "Hyperventilate" or "hyperdrive" won't trigger it.

• Something has to follow it. A bare "Hyper." with no instruction is transcribed normally.

Tip: Speaking a command and nothing appears? Turn on custom vocabulary—biasing the recognizer toward the spelling Hyper makes your model emit it cleanly instead of "hyper" look-alikes.

Code mode

End the wake word with code—say "Hyper code …"—and Hyprcore switches to code output. It returns raw code with no prose, no explanation, and no markdown fences, ready to paste straight into your editor.

• "Hyper code a debounce function in TypeScript" → the function, nothing else.

• "Hyper code fix the off-by-one here" (with code selected) → the corrected code.

Turn it on

Action Mode lives in Settings → Dictation, under Behaviour. Toggle Action Mode on. It controls the action_mode_enabled setting.

Because Action Mode routes through an LLM, make sure you've configured a post-processing provider first.

Add your own wake words

"Hyper" is always available, but you can add your own triggers—including a friendlier "Hey Hyprcore"—via the action_mode_wake_words setting. Your custom words are matched alongside the built-in "Hyper" family, so adding one never disables the default.

Two things to know:

• A custom wake word ending in code selects code mode, exactly like "Hyper code". Add "computer code" and "computer code a bash loop" returns raw code.

• Longer wake words win over shorter ones, so a two-word trigger is matched before a one-word prefix of it.

What's next

• Post-processing — Pick the LLM that runs your Action Mode instructions.

• Custom vocabulary — Bias the recognizer so "Hyper" and your own terms come through cleanly.