heyski.io / tutorials / voice coding
Tutorial 01 · Start talking

Voice coding: talk to your coding agent

From opening the widget to a spoken conversation with Claude Code, Cursor, Codex — any agent — in about a minute. Everything runs on your machine.

1:22 · with sound · everything shown runs on-device
First, the term

What is voice coding, exactly?

Voice coding means writing software by speaking instead of typing — and the term now covers two very different generations of tools.

The first generation is code dictation: grammar-based tools like Talon and Serenade, built first for developers with RSI. You speak symbols, syntax, and editor commands, and the tool types them for you. Genuinely powerful — and you are still the one programming, character by character, with a command grammar to learn.

The second generation is talking to an AI coding agent. With an agent like Claude Code, Cursor, or Codex in your project, you don’t dictate code — you describe outcomes. The agent reads the files, writes the code, runs the tests; you speak at conversation speed and hear it answer. Voice coding stops being transcription of the work and becomes a conversation about it.

SKI is built for that second kind — a two-way spoken loop with your agent, running entirely on your own machine.

Code dictation (Talon, Serenade…)Voice coding with an agent (SKI)
You saySymbols, syntax, editor commandsWhat you want, in plain language
Who writes the codeYou — spoken keystroke by keystrokeThe agent — it reads, writes, and runs tests
It answers backNo — one directionYes — out loud, in a natural voice
Learning curveA command grammar to memorizeNone — just talk
Best forPrecise hands-free editingBuilding, refactoring, and reviewing with an agent

Both kinds are real voice coding, and the dictation lineage matters — especially for accessibility. They just solve different problems, and they compose: nothing stops you from running both.

The written version

Five steps, no typing.

STEP 01

Open SKI

Launch SKI and a small widget appears on your desktop — that's the whole interface. On Mac it can dock into the notch or float as a pill; on Windows it floats as a pill wherever you park it.

Infographic: start SKI in three steps — open your coding agent, type ski, then just speak
The whole flow on one card
STEP 02

Type ski in your agent

In your coding agent — Claude Code, Codex, any of them — type ski. That connects the session to the widget. When the dot turns green, you're connected: your agent can now hear you and speak back.

STEP 03

Connect as many projects as you like

It's not just one project. Connect several at once — even from different agents — and jump between them. Click the project name on the widget to switch; each project answers in its own voice, so you always know who's talking.

STEP 04

Just speak

Say what you need out loud. Your words land in the agent as text and it gets to work — reading files, writing code, running tests. And when it's done, it talks back, out loud, in a natural voice generated on your machine. You talk. It works. It answers.

STEP 05

The quick controls

Hover the widget for the controls that matter mid-flow: mute (mic off at the source), silent mode (replies as text only), and screenshots that ride along with your next spoken sentence. Every one of them can be a hotkey — yours to choose, in Preferences → Hotkeys.

Infographic: the SKI pill explained — mark color shows state, hover reveals settings, meeting menu, silent, mute and quit
The widget, explained
Infographic: show SKI your screen — ask the agent to take a screenshot, or press your own screenshot hotkey
Screenshots ride your next sentence
Agent by agent

Voice coding with your agent.

Claude Code

Voice coding with Claude Code

Talk to Claude Code out loud — in the terminal or your IDE — and hear it answer.

Read the guide →
Cursor

Voice coding in Cursor

Speak to Cursor's agent while you stay in the editor — it works, then talks back.

Read the guide →
Codex

Voice coding with Codex

Give OpenAI's Codex a voice — spoken requests in, spoken answers out.

Read the guide →
Voice coding · FAQ

Good questions.

What is voice coding?

Voice coding is writing software by speaking instead of typing. Historically it meant dictating code word by word with grammar-based tools; with AI coding agents it means describing what you want out loud while the agent writes, runs, and tests the code. SKI is built for the second kind — you talk to your coding agent and hear it answer.

Is voice coding faster than typing?

Dictating code symbol by symbol usually isn't. Directing an agent by voice is — you speak at around 150 words a minute, several times faster than most people type prompts, and the agent does the actual typing. The speed comes from delegation, not transcription.

Do I need to learn voice commands or a grammar?

No. There is no command language to memorize — you speak normally, and your agent interprets it exactly as it would a typed message. Saying what you want is the whole interface.

Which coding agents can I talk to?

Claude Code, Codex, Cursor, Gemini CLI, Windsurf, and OpenClaw — SKI installs a small skill for each with one click, and you can connect several projects from different agents at the same time.

Does SKI type into my editor or terminal?

No. SKI talks to your agent through a skill and plain files inside your own project — nothing is injected into windows or keystrokes. Your agent reads what you said and replies through the same files, which is why it works with any terminal, editor, or IDE.

Can I review what I said before it's sent?

Yes — turn on approve-before-send and every transcript lands in an editable bubble first. Nothing reaches the agent until you confirm it.

Is there push-to-talk?

On Mac, yes — opt in to the globe/fn key: tap it to mute or unmute, or hold it to talk. You can also bind your own hotkeys for mute, silent mode, and screenshots in Preferences → Hotkeys.

Does voice coding need an internet connection?

No. Speech recognition and the voice that answers both run on your own machine — the whole loop works offline, and nothing you say is uploaded.

Keep going

More tutorials.

Tutorial 02

Record any meeting — free, on-device

No bot joins, nothing uploads, no time limits. Live transcript, then your agent writes the notes.

Watch & read →
Tutorial 03

Send your agent to the meeting

Paste a link and your agent joins the call — speaking live, or silently taking notes.

Watch & read →

Try it out loud.

Free for life. Register once — no card — and the whole voice loop is yours, on-device, forever.

Download for Mac Download for Windows Download for Linux
macOS 14.4+ · Apple Silicon  ·  Windows 11 & 10 · x64  ·  Linux · deb · x64