Short answer: click Lovable's prompt field, press ⌃+⌥+R (Mac) or Ctrl+Alt+R (Windows/Linux), and dictate a complete app-builder spec: pages, Supabase tables, auth method, RLS rules, API routes, and acceptance criteria. AICHE inserts text. Lovable generates the app. AICHE does not edit components or deploy.
The Problem
Lovable quality tracks prompt completeness: React pages, Tailwind layout, Supabase schema, auth provider, Stripe webhook, GitHub sync target. Typed prompts stop at "task manager with login" because 300 words of schema detail is tedious at 40 WPM.
Dictate the full builder prompt once: table names, column types, row-level security, /dashboard route behavior, empty-state copy, and verification steps ("seed two users, show admin-only settings").
What Changes
Speaking runs at around 150 WPM. The same 300-word prompt that takes 7 minutes to type takes about 2 minutes to say. More importantly, speaking lets you think out loud without editing in real time. You describe the dashboard layout, remember the authentication edge case, add the Stripe integration requirement, describe the mobile breakpoint behavior - all in one continuous flow. Nothing gets trimmed because typing it felt like too much work.
AICHE captures the speech, removes filler words and false starts, and inserts clean text into Lovable's prompt field. You review, hit Enter, and Lovable builds.
How It Works
- Open lovable.dev and create a new project (or open an existing one).
- Click into the prompt field.
- Press ⌃+⌥+R (Mac) or Ctrl+Alt+R (Windows/Linux) to start recording.
- Speak the full requirement - describe the UI, the data model, the backend behavior, the user flows.
- Press the hotkey again. AICHE transcribes, cleans up filler, and inserts the text.
- Review the prompt, hit Enter.
- Lovable's AI generates the code.
The hotkey works globally. You don't need to click anything to activate it. AICHE uses Smart Insert to place text wherever the cursor is, including inside browser text fields like Lovable's prompt box.
Prompting Lovable With Voice
Lovable has three main modes you'll use regularly:
Default chat mode handles most single-step tasks. Describe what you want built and Lovable generates it. Voice works well here because you can describe a complete component - layout, interactivity, data source, edge cases - without stopping to abbreviate.
Agent Mode is for complex multi-step builds. Lovable acts more autonomously here: it explores the codebase, plans before coding, and debugs itself. Lovable reports Agent Mode reduces build errors by around 90% versus single-shot generation. The prompts going into Agent Mode tend to be longer because you're setting the full scope up front. Speaking a 400-word architectural brief is faster and more complete than typing a 100-word summary that forces Lovable to guess.
Plan Mode is iterative and collaborative - you describe changes, review the plan before Lovable executes, and steer at each step. Voice fits naturally here too: describe a revision out loud, review what Lovable proposes, speak the next adjustment.
What to Include in a Spoken Lovable Prompt
Lovable generates better code when prompts cover:
- The tech stack (React, TypeScript, Tailwind are the defaults - confirm or override)
- Authentication method (Lovable Cloud, Supabase auth, JWT)
- Data model (table names, key fields, relationships)
- Page-by-page layout requirements
- Business logic rules (what triggers what, who can do what)
- Integrations needed (Stripe, Resend, custom APIs, MCP connectors)
- Deployment target (Lovable Cloud, Vercel, custom domain)
Speaking through all of this takes 2-3 minutes. Typing it takes 15-20. The completeness difference directly affects how many follow-up prompts you need.
Iterating on an Existing Lovable Project
Voice is also useful mid-project, not just for the initial build. Lovable's prompt field is always there as you iterate. Common scenarios:
You're reviewing the generated app and notice missing behavior. Instead of typing out a full correction, click into the prompt, speak the issue and the fix, and let Lovable apply it. The spoken description is usually more precise than what you'd type because you're not trying to be brief.
You want to add a feature that touches multiple parts of the app - a new user role with different permissions, a notification system, an export function. Speak through the full scope of what changes. Lovable's codebase awareness means a thorough description lands cleanly.
You need to debug something. Describe what you're seeing, what you expected, and any context about the relevant component. Lovable uses that context to search the codebase and fix the right thing.
GitHub Sync and the Handoff
Lovable syncs code to GitHub from the start, so if you need to move into a real IDE at some point, the codebase is already there. For teams using Claude Code, Cursor, or another AI coding tool alongside Lovable, the handoff is straightforward - the code is version-controlled, readable, and editable outside Lovable.
AICHE works the same way in those tools. The workflow doesn't change: hotkey, speak, insert. Whether you're prompting Lovable to build or prompting Cursor to refine, the voice input layer is the same.
Common Questions
Q: Does AICHE work in browser-based apps like Lovable?
A: Yes. AICHE inserts text into whatever field has the cursor, including text inputs inside browser tabs. Click into the Lovable prompt field, use the hotkey, and AICHE drops the transcribed text there.
Q: Lovable prompts often include technical terms - API names, component names, database fields. Will those transcribe correctly?
A: For common framework names (React, Supabase, Tailwind, Stripe, Resend) transcription is accurate. For project-specific names - your table names, internal service names, custom API endpoints - add them to AICHE's custom vocabulary. They'll spell correctly every time after that.
Q: Can I dictate mid-session while Lovable is generating?
A: Yes. Lovable queues prompts. You can dictate and insert a follow-up prompt while the current generation is still running.
Q: My initial prompts keep producing apps that need heavy iteration. Would longer voice prompts actually help?
A: Yes, and it's the core of Lovable's prompting advice. The platform's own guide says detailed, structured prompts reduce errors significantly. If you've been typing abbreviated prompts to save time, voice removes that constraint. Describe everything you know up front - layout, data, logic, edge cases - and let Lovable handle more in the first pass.
Q: What about non-English speakers building in Lovable?
A: Turn on Auto-translation in AICHE settings. Speak in your native language and AICHE outputs English. Lovable's prompt box receives clean English regardless of what language you think in.
Result: the prompts you've been shortcutting because typing them felt like too much work arrive complete and in seconds. Lovable builds more accurately, iteration cycles drop, and the main bottleneck shifts to reviewing the output rather than writing the input.
Try it now: open Lovable, start a new project, press your hotkey, and speak one complete app idea - describe the main screens, what data it stores, how users authenticate, and one key business rule. Hand the full prompt to Lovable and see what a thorough first pass produces.