Voice mode for AI coding: talk through the hard parts

Voice mode for AI coding is not a gimmick mic button. It is a second input for the same multi-model chat: explain a bug, walk a refactor, or dictate a research angle while the thread — and the model picker — stay intact.

When voice helps coding

• Rubber-ducking a stuck design before you touch the keyboard
• Narrating stack traces and error context hands-free
• Long planning sessions where typing slows thinking
• Pairing voice with a frontier model for hard reasoning, free models for volume

When the job is pure paste-and-patch, stay on text. Use voice when speech is the faster interface to the same workspace.

How it works on zy.ai

• Ultra only — Voice mode is included on Ultra ($45/mo)
• Continuous listen style with Esc to exit (ambient presence UI)
• Same conversation history as typed chat
• Still multi-model: Free models on Free; frontier on Pro/Ultra; Voice itself is Ultra

Plans and product rules: /docs. Upgrade ladder: Free vs Pro when to upgrade.

A practical voice coding loop

1. Open the thread with the error or goal already pasted.
2. Enter Voice; describe what you tried and what “done” means.
3. Ask for a plan first, then a minimal patch.
4. Switch models mid-thread if another engine is better for the next step — Claude vs GPT vs Gemini.
5. Exit Voice, apply the diff, verify with tests.

Voice is for thinking speed. Diffs still need your eyes.

Voice vs agents vs MCP

• Voice — speak to multi-model chat (Ultra)
• Business agents — SEO, company, leads research (Ultra)
• MCP tools — GitHub, Notion, etc. on Free/Pro/Ultra member accounts

MCP overview: MCP tools in AI chat. Agents: business search agents.

Who should not buy Ultra only for Voice

If you never talk through problems and your work is short typed prompts, Free or Pro is enough. Buy Ultra when capacity, Voice, or research agents show up every week — not for the badge. Decision help: when free AI is enough.