Coding by Voice: How to Use Claude Code From Your Phone

What is coding by voice?

Coding by voice is the practice of writing and changing software by speaking instead of typing. The term has been around for years, but for most of that time it meant literal dictation — narrating every keyword, symbol, and line break to a speech-to-text engine. It was slow, brittle, and mostly used out of necessity.

The arrival of AI coding agents changed the definition. Today, voice coding means describing your intent in plain language and letting an agent translate that intent into actual code changes. You don't say "open paren, i equals zero, semicolon" — you say "add pagination to the users endpoint," and the agent does it.

Dictation vs. an autonomous coding agent

The distinction matters for search and for reality:

• Voice dictation transcribes your words into text verbatim. Useful for comments or commit messages, useless for real logic.
• An AI coding agent like Claude Code takes a spoken instruction, reasons about your codebase, edits files, runs commands, and reports back. This is what makes hands-free coding genuinely productive.

Tulpar is built entirely around the second model: your voice is the interface, and Claude Code is the engine doing the work.

Why code by voice?

Typing isn't going away — but a surprising amount of engineering work doesn't need a keyboard at all. Here's why developers are adopting hands-free coding:

• Code from anywhere. Kick off a fix on a walk, review a pull request on the train, or start a refactor from the couch. Running Claude Code on your phone untethers you from your desk.
• Lower friction to start. Speaking a task takes seconds. That low activation energy means you actually capture ideas instead of losing them before you're back at your machine.
• Parallelism. Because each conversation runs in isolation, you can dictate three tasks across three projects and let them run at once.
• Accessibility. Voice-first coding reduces repetitive strain and opens programming to people for whom a keyboard is painful or impractical.
• Flow. Talking through a problem is a natural way to think. Explaining what you want out loud often produces a clearer spec than typing it.

How Tulpar turns speech into shipped code

Under the hood, coding by voice with Tulpar is a short, reliable pipeline:

1. Speak. You tap the mic on your phone and describe a task in plain English.
2. Transcribe. High-accuracy speech-to-text turns even long, rambling thoughts into a clean prompt.
3. Act. Claude Code runs on your own server, working inside an isolated git worktree so it can edit files and run commands without touching your main branch or other sessions.
4. Hear back. The agent speaks a summary of what it did and shows you the full diff to review.

How to start coding by voice with Claude Code

Getting set up for voice to code takes only a few steps:

1. Run the server. Tulpar hosts Claude Code on a small server you control (a cheap cloud VM is plenty).
2. Open the web app on your phone. No app store needed — it's a progressive web app. On iOS you can add it to your home screen so it launches full-screen like a native app.
3. Create a project. Point it at a git repository and Tulpar sets up an isolated worktree for each chat.
4. Tap the mic and talk. Describe your first task and let the agent work. Review the diff, then ship.

Real voice coding workflows

Coding by voice isn't a gimmick when the agent is strong. A few workflows where it genuinely beats waiting until you're back at a keyboard:

• Fix-it-on-a-walk. A bug report comes in. You describe the symptom, the agent locates and patches it, and you review the diff before you're home.
• Parallel kickoffs. Start a dependency upgrade in one project, a docs pass in another, and a failing-test investigation in a third — all running at once.
• Spoken code review. Ask the agent to summarize a diff, then dictate the changes you want. Great for reviewing on the go.
• Rubber-duck planning. Talk through an approach out loud; the agent turns the conversation into a concrete plan and the first commit.

Tips for effective voice coding

• Speak intent, not syntax. Say what outcome you want and let the agent choose the implementation.
• Give it context once. Mention the file, feature, or endpoint by name; the agent remembers it for the rest of the chat.
• Always review the diff. Voice makes starting work effortless — but you're still the one who ships. Read the change before merging.
• Use separate chats for separate tasks. Isolation keeps parallel work from colliding and keeps each conversation focused.

Limitations to know

Honest expectations make hands-free coding more useful:

• It's best for describing and directing work, not for pixel-level edits — precise cursor work is still easier at a keyboard.
• You need a quiet-enough environment for accurate transcription.
• Reviewing large diffs on a phone screen has limits; save big reviews for a larger display.

Frequently asked questions

Can you really code by voice?

Yes. With an AI coding agent like Claude Code, you speak your intent in plain English rather than dictating syntax. The agent writes, edits, and runs the code for you, so voice coding is practical for real work — not just note-taking.

Can I use Claude Code on my phone?

Yes. Tulpar runs Claude Code on your own server and gives you a mobile web app to drive it. You speak a task from your phone, the agent works in an isolated git worktree, and it speaks the result back. You can add it to your iOS home screen as a full-screen app.

Is coding by voice faster than typing?

For describing intent, speaking is often faster than typing — and you can do it away from a keyboard. Because the agent handles the actual code, voice coding shines for kicking off tasks, reviews, and fixes while you're mobile rather than for character-by-character editing.

Is voice coding good for accessibility?

Yes. Hands-free coding can reduce strain from repetitive typing and helps developers who find a keyboard difficult. Speaking intent to an AI agent lowers the physical barrier to writing and shipping code.