Skip to content

Voice typing · macOS · Windows

Speak. Ship. Move on.

Hold a hotkey, speak naturally, and watch polished text land in the app you’re already in — fillers gone, punctuation right, terms preserved.

Download for macOS Download for Windows

Free during beta · unsigned build, one-time Gatekeeper bypass

How it works

Hold. Speak. Done.

Three keys, one habit. No window to switch, no panel to open, no model to pick. Your cursor stays where it is.

  1. 01

    Press and hold the hotkey

    Cmd + Alt + Space anywhere, or hold Fn / Right Ctrl as a one-finger modifier. Your choice.

  2. 02

    Talk like you’d talk to a person

    “Move the launch to Friday and tell Anna we’re a day late.” Pause where it’s natural. Don’t dictate punctuation.

  3. 03

    Polished text appears at your cursor

    Fillers stripped, punctuation added, your dictionary terms spelled the way you want. Then it pastes.

What it does

Built for the way you already work.

Not a notepad. Not a chatbot. A dictation layer that respects whatever app has your focus.

Push-to-talk dictation

Hold a key, speak, release. Works in any app — Slack, Linear, your editor, the terminal. Combo or single-modifier, you choose.

Command Mode

Select text in any app, hold Fn + Ctrl, say “make this professional” or “translate to Russian.” The selection rewrites in place.

Polish that sounds like you

Optional LLM cleanup removes “ums”, fixes punctuation and casing, and preserves your voice — never paraphrases the meaning.

Personal dictionary

Names, acronyms, product terms — 3-layer injection: speech biasing + LLM prompt + post-process. Your terms always spell right.

Smart suggestions

When you fix the same word twice, Iskra notices and offers to add it to your dictionary. Quietly, never in the way.

Searchable history

Last 500 transcripts kept locally, virtualized search, one-click re-insert. Audio retention policy is yours to set.

Command mode

Edit selections with your voice.

Select anything — an email draft, a Linear ticket, a paragraph in code comments — and rewrite it without lifting your hands.

  • Translate to Russian
  • Shorter, bullet points
  • Past tense
  • Sound less defensive
  • Add a closing line
Fn Ctrl · say “Make this professional.”
Selected draft

hey so um we need to push the launch back to friday I think — there's a bug in the build pipeline that anna is looking into rn

“Make this professional.”
Rewritten pasted

We need to push the launch to Friday — there’s a bug in the build pipeline that Anna is investigating right now.

A quiet panel. Loud results.

One window, four screens. No model picker, no provider switcher, no “advanced settings” drawer. Defaults work.

Iskra
Ready · Cmd+Alt+Space
  • 2 min ago · Slack

    Pushed the migration script to staging — Anna, can you take a look before noon?

  • 11 min ago · Linear

    Spec: voice-driven inline edit. Selection + hotkey → rewrite-in-place via LLM.

  • today · Cursor

    TODO: collapse the legacy auth middleware once the new gate is wired up.

Polish

ok so uh I think we should um basically move the deploy to like tomorrow morning

I think we should move the deploy to tomorrow morning.

Dictionary

  • OpenRouter ← open router, openrooter
  • Forgejo ← forge yo, forge joe
  • Tauri ← tower, tory, tauree
Recording · 0:04 · Esc

Privacy

No surprises in your audio or your text.

Audio goes through one provider only — OpenRouter — and nothing else. Keys live in your OS keychain. History stays on your machine.

  • One provider, one key

    OpenRouter handles speech-to-text and optional LLM polish. No secondary model picker, no “fallback” fanout. Your key, your choice of model.

  • Keys in the OS keychain

    macOS Keychain or Windows Credential Manager. Iskra never writes your key to a settings file or to any log line.

  • History is local-only

    SQLite on your machine. Audio retention is a slider — keep none, the last day, or the last 500. Iskra never uploads transcripts back.

  • Permissions, explained

    Mic for capture · Accessibility for paste · Input Monitoring for hold-to-talk. We deep-link to the System Settings panes so you can audit.

Install

v0.1.3

60 seconds, then it’s yours.

Builds are unsigned during beta. macOS Gatekeeper and Windows SmartScreen will ask once — after that, in-app updates land silently.

Build from source

Go + Rust + React. One make sidecar, one pnpm tauri build.

Read the build doc

Frequently asked

The questions that come up.

  • Does Iskra work offline?

    Not today. Speech-to-text and LLM polish go through OpenRouter. A local-STT path is on the roadmap for after beta — until then, you need a network connection.

  • Where does my audio go?

    Straight to OpenRouter using your API key. Iskra doesn’t proxy through any server we control. We don’t see your audio, your transcripts, or your dictionary.

  • Which languages does it understand?

    Whatever OpenRouter’s Whisper-class models understand — 90+ languages. Polish prompts are tuned for English and Russian today; other languages still get cleaned, just less aggressively.

  • Why is it unsigned?

    Apple Developer and Microsoft Trusted Signing cost money and we’re solo-funded during beta. You bypass Gatekeeper / SmartScreen once and the in-app updater takes it from there.

  • What does it cost?

    Iskra is free during beta. After GA there’ll be a small per-month plan; you’ll still bring your own OpenRouter key, so the cost of speech doesn’t change.

  • How is this different from Whisper inside VS Code?

    Iskra is a global hotkey, not an editor plugin. It works in your browser, your terminal, your design tool, your Slack thread — anywhere a cursor lives.