Why bother?

Because sometimes you're pacing around the room thinking through a problem and stopping to type breaks the flow. Or you're on the sofa and your laptop is over there. Or, honestly, because your wrists hurt and you've been typing since 6am. Voice coding isn't about replacing the keyboard. It's about having another input when the keyboard isn't ideal.

The tooling went from "barely works" to "actually useful" around late 2025 when Deepgram and Whisper got good enough for technical speech. Three approaches, going from brilliant and free to polished and paid.

justspeaktoit

justspeaktoit by Chris Mitchelmore is the one. Absolute belter. It's a macOS menu bar app that pipes speech straight into whatever's focused. Uses Deepgram for transcription, which gives you $200 free credit to start with. That's a lot of talking.

Here's why I rate it: Wispr Flow does roughly the same job for 9 quid a month. justspeaktoit does it for free (until you burn through the Deepgram credit, which takes ages). You don't even need a post-processor. Deepgram's raw output is clean enough for code instructions. The speed is brilliant too, fast enough that you don't notice the gap between speaking and text appearing. Way quicker than waiting for a local model.

There's an iOS app in there as well, which is a nice touch. The era of building your own tools for everything is properly here. Chris just knocked this out and it works better than commercial alternatives costing real money.

Parler

Parler is a Tauri desktop app (Rust + React) that does offline speech-to-text. Press a hotkey, speak, text appears wherever your cursor is. macOS, Windows, Linux.

Offline is the word that matters. Runs Whisper models locally, the newer Parakeet V3 works on CPU with no GPU needed. Nothing leaves your machine. No subscription. No "200 free minutes then pay up." Just a binary that turns speech into text.

It's got push-to-talk and toggle modes, voice activity detection to filter silence, and it's explicitly designed to be forkable. The maintainer's line is "accessibility tooling belongs in everyone's hands, not behind a paywall." Hard to argue with that.

For agent work, you speak your intent and it types it out. Pipe that into Claude Code's input and you've got voice-driven agentic coding for free. Not fancy, but it works. I reckon this is where most people should start.

Wispr Flow vs the free alternatives

Wispr Flow is the polished commercial option at about 9 quid a month. Proper good dictation, understands code terms, works system-wide on Mac. If you want something that just works out of the box and you don't mind paying, it's the benchmark.

The Deepgram-based approaches give you 200 free hours which is generous. Good accuracy on technical speech. The gap between free and paid has closed a lot since mid-2025.

Where this is going

Anthropic shipped voice mode in Claude Code in March 2026. That's the signal. The standalone tools are filling a gap until the platforms catch up, same story as remote coding. Parler if you want free and offline, Wispr if you want polished and paid, justspeaktoit if you want the best balance of both.

For the more ambitious idea of actually calling your agent on the phone, see claude-phone and voice call experiments.