01
Record
Microphone or file. MediaRecorder API directly in the browser, no uploads.
Client-only · BYOK · Apache-2.0
Speechto AIPrompt.
Scribably turns speech into prompts. Runs in your browser. No server, no account. Your keys, your mic.
- 0
- Server roundtrips
- 4
- Providers included
- Apache 2
- Open Source
Rec · 00:12
Transcript
Write me an email to the team, keep it short, friendly tone, project kickoff on Monday
groq · whisper-large-v3en-US
Your keys. Your browser.
How it works
Four steps between "uh…" and a finished prompt.
Each step is configurable. You decide where your audio goes. And where it stops.
02
Transcribe
Pick your STT provider: Groq Whisper or OpenAI Whisper. Language detection included.
03
Clean
LLM removes filler words, fixes sentence structure and grammar. Optional, toggleable.
04
Prompt
Second step: cleaned text becomes a structured prompt. Copy, export, Markdown, LaTeX.
Privacy by Design
Your voice leaves your browser only where you send it.
Scribably is a pure client app. No Scribably server. Nothing logs or forwards anything. API keys stay in your
localStorage, audio goes directly from you to your provider.- Try it now — no signup, no account. Bring your own key for unlimited use.
- localStorage only — nothing on a backend
- Open source — code is on GitHub
- Apache-2.0 — fork it, self-host it, take it apart
~ / scribablylocalStorage
const keys = {
stt: 'sk-...',
llm: 'sk-ant-...',
}
// stays here. period.
localStorage.setItem('wp-config', …)
Zero Server
Provider-agnostic
Four providers, mix and match.
Two independent provider registries: one for Speech-to-Text, one for Text Processing. Mix and match. New provider? One factory function, done.
- STT: Groq Whisper, OpenAI Whisper
- LLM: OpenAI-compatible, Anthropic
- Two-step pipeline: Clean → Prompt
- Connection test for every key before sending
STT + LLM
Groq
Whisper Large V3. Fast, cheap, default.
STT + LLM
OpenAI
Whisper API · GPT-4o-compatible endpoint.
LLM
Anthropic
Claude Sonnet / Opus. Direct from your browser.
LLM
OpenAI-compatible
OpenRouter, LMStudio, Ollama, your own server.
Two-step pipeline
From "um, well, yeah" to a structured prompt.
Raw text is rarely what you want to send. Scribably splits cleanup and prompt construction into two toggleable steps. Grab the output at any point.
- Step 1: Cleanup — filler words, sentence structure, typos
- Step 2: Prompt — structured, role-based, instructive
- Both steps triggered manually, nothing automatic
- Custom system prompts per step — your recipes
Raw00:12
um, uh, can you write an email to the team saying we kick off monday, not too formal but not too stiff either you know
cleanup + prompt
Promptready
Write a warm, professional email to the team. Content: project starts Monday, tone: collegial, not too formal. Max. 6 sentences.
Export & Mobile
Out of the browser. Into every corner.
Copy-paste, Markdown, LaTeX, QR transfer to your phone. Dark mode too.
- Export: Clipboard · .txt · .md · LaTeX
- QR transfer: config as Base64 QR to your phone
- Mobile-first responsive UI
- Dark mode for late nights
Transfer · Config
Base64encrypted localhost
Scan → Setup
Built with
React·TypeScript·Vite·Tailwind 4·Vitest
FAQ
Common questions
- What is Scribably?
- Scribably is a free, open-source web app that converts speech into structured AI prompts. It runs entirely in your browser — no server, no account, nothing stored remotely. You dictate what you need; Scribably transcribes the audio, cleans up the raw text, and turns it into a ready-to-use prompt for your AI model of choice.
- How does Scribably protect my privacy?
- Scribably has no backend. Your API keys are stored only in your browser's localStorage and are never sent to Scribably infrastructure. Audio travels directly from your microphone to your chosen provider (Groq or OpenAI) — not through any Scribably server. Analytics are handled by self-hosted Umami, which sets no cookies and stores no personal data. The full source code is on GitHub and auditable.
- Which AI providers does Scribably support?
- For speech-to-text, Scribably supports Groq Whisper and OpenAI Whisper. For text processing and prompt construction, it supports any OpenAI-compatible API and Anthropic Claude. The two provider registries (STT and LLM) are independent — you can mix providers freely, for example using Groq for fast transcription and Anthropic for prompt construction.
- What does BYOK mean — do I need an API key?
- BYOK stands for Bring Your Own Key. Scribably connects directly to your Groq, OpenAI, or Anthropic account — there is no shared API access or hidden subscription. Your key, your usage, your costs. Groq offers a generous free tier, so you can get started without spending anything.
- Is Scribably free to use?
- Yes. Scribably is free and open source under the Apache-2.0 license — free to use, fork, self-host, and modify. Provider costs (for transcription and LLM calls) depend on your usage and the pricing of your chosen AI provider, not Scribably itself.