Talk to Aider by Voice: Hands-Free Pair Programming
Aider turns your terminal into an AI pair programmer that edits your codebase directly. The one friction point is that you still type every instruction by hand. Dictating those prompts by voice makes the loop faster and, done right, keeps your speech entirely on your Mac.
Key takeaways
- Aider reads plain text from the terminal, so any real dictation tool can feed it your voice.
- On-device speech-to-text keeps your spoken prompts private, even though Aider still calls a model.
- A custom dictionary handles function names, library names and project jargon.
- Most people speak around three to four times faster than they type, which suits long coding prompts.
Why voice fits Aider so well
Aider is a command-line tool: you type an instruction, it edits files, shows a diff, and commits. The prompts you write are usually full descriptive sentences, like "refactor the auth handler to use the new token helper and add a test for the expired case." That is exactly the kind of long, natural-language request that is slow to type and quick to say. According to the general research on typing and speaking rates, most people speak around three to four times faster than they type, and you can read more on words per minute if you want the background.
Voice also keeps you in a thinking posture. Instead of switching to typing mode, you describe intent out loud, glance at the diff, then dictate the next tweak. It is the closest a terminal tool gets to talking with a real pair programmer. The same approach works for other terminal assistants too, whether you dictate prompts to the Gemini CLI or drive a coding agent in an editor.
How the hands-free loop works
The setup is simple because Aider does not need a special integration. It reads whatever text arrives at its prompt, so dictation just needs to type at your cursor. Here is the flow from spoken word to committed change.
Notice that two of the four stages run on your Mac. With BlaBlaType, both the speech recognition and the AI cleanup happen on-device, so your spoken prompt is never uploaded. Aider itself still sends the finished prompt to whatever model you have configured. If you want the whole loop private, point Aider at a local model, the same way you would talk to a local LLM by voice with Ollama.
Setting it up on a Mac
You do not need to change how you run Aider. You are adding a dictation layer on top of the terminal you already use.
- Install a system-wide dictation app. It has to type at the cursor in any window, including your terminal, not just into a notes app.
- Pick a trigger shortcut. A single hotkey to start and stop recording means your hands leave the keyboard only when you want them to.
- Add your jargon to the dictionary. Put function names, package names and product terms in the custom dictionary so they transcribe correctly.
- Speak your prompt, then review. Dictate the instruction at the Aider prompt, press enter, and read the diff before you accept it.
The custom dictionary is the piece developers underestimate. Generic dictation will mangle a name like useAuthToken or psycopg, but once those live in the dictionary they land correctly every time. On-device AI cleanup handles the rest: it strips "um" and "you know", fixes punctuation, and turns a rambling sentence into a tidy prompt before Aider ever sees it.
On-device vs cloud dictation for coding
Not every dictation tool is built the same way, and for source code the difference matters. Here is how the common approaches compare for a terminal workflow.
| Approach | Voice stays local | Types in terminal | Custom jargon | Works offline |
|---|---|---|---|---|
| BlaBlaType (on-device) | Yes | Yes | Yes | Yes |
| Cloud dictation service | No | Yes | Varies | No |
| Built-in Mac dictation | Mixed | Yes | Limited | Mixed |
| File transcription tool | Yes | Files only | No | Yes |
For a hands-free Aider loop you want the top row: voice that stays on the machine, types straight into the terminal, and learns your vocabulary. Cloud tools upload every prompt, which is a hard no for many teams working under an NDA. If you are weighing local transcription engines more broadly, our comparison of Whisper apps, MacWhisper vs Superwhisper covers the trade-offs. The same voice layer also lets you talk to ChatGPT by voice when you switch from the terminal to a chat window, and it plugs into editor agents documented in the Cursor docs too.
Pair-program with Aider, hands-free
Dictate prompts straight into your terminal with 100% on-device speech-to-text. No card needed for the trial.
Download for macOSGetting the most out of voice prompts
A few habits make dictated Aider prompts noticeably better. Speak in complete instructions rather than fragments, because the AI cleanup and Aider both work better with a clear sentence. Name files and functions explicitly, since spoken references like "that thing we changed" give the model nothing to anchor to. And keep prompts scoped to one change at a time so the diff stays reviewable. BlaBlaType also supports custom AI prompts, so you can set a house style, for example asking it to phrase everything as an imperative instruction, which reads cleanly at a coding prompt.
If you dictate across many tools during the day, from your terminal to your notes to a browser AI chat, it helps to standardize on one dictation setup rather than juggling several. For a wider view of the options, see the roundup of the best dictation software for Mac, and check the pricing page for the plans that unlock features like file transcription and screen-context awareness.
Frequently asked questions
Can I use Aider by voice on a Mac?
Yes. Aider reads text from your terminal, so any system-wide dictation tool can feed it your voice. With BlaBlaType you press a shortcut, speak your instruction, and the transcribed text lands at the Aider prompt, no copy and paste needed.
Does dictating prompts to Aider send my code to the cloud?
The dictation step does not have to. BlaBlaType transcribes your speech 100% on-device, so your spoken prompt never leaves your Mac. Aider itself still sends prompts to whichever model you configure, so choose a local model if you want a fully private loop.
Is voice faster than typing prompts to Aider?
For most people, yes. Most people speak around three to four times faster than they type, and coding prompts are often long, descriptive sentences that are quicker to say out loud than to type by hand.
Will dictation get my variable names and jargon right?
BlaBlaType has a custom dictionary where you add function names, library names, and project jargon so they transcribe correctly. On-device AI cleanup also fixes punctuation and removes filler words before the text reaches Aider.
Do I need to keep my hands on the keyboard?
No. That is the point of a global shortcut. You trigger dictation with one key, describe the change out loud, and review the diff. Your hands stay off the keyboard for the prompt-writing part of the loop.