Why use saymd instead of built-in voice in Cursor, Claude Code or Copilot?

Built-in voice turns speech into chat input inside one tool. saymd turns speech into a structured Markdown prompt file in your repo — Objective, Context, Instructions, Constraints — that any agent can read and that you can version, share, and reuse.

Free MIT CLI · BYOK · macOS + Linux · Windows via WSL

Speak once. Get a prompt file any agent can @.

Agents execute. saymd delivers a usable spec — structured Markdown from messy speech, ready to @ in any AI coding agent.

Get Free on GitHub Get Pro — €39/year

Free CLI is MIT · Pro is a separate commercial package · €5/month available

Keys stay local. Audio goes only to the STT provider you chose — not to saymd.app.

terminal

npx saymd setup
npx saymd
npx saymd --file idea.m4a --template feature

From messy speech to an agent-ready spec.

Speak for about 30–60 seconds — or pass an existing recording with --file. saymd writes a reusable Markdown prompt instead of trapping the idea inside one AI tool.

Works with: Cursor · Claude Code · Codex · Grok Build · GitHub Copilot · Antigravity CLI · OpenCode

Codex / ChatGPT plugin Grok Build plugin Antigravity CLI plugin

Install — npx saymd setup
Speak — mic or --file; get .ai/prompt.md
@ the prompt — open it in your agent

--template feature · from the mic

You say

“Okay, for saymd I need Google OAuth on the Next.js App Router — users should land in a normal session cookie after Google. Rate-limit the auth callback so we don’t get hammered. No password storage at all, and don’t put refresh tokens in localStorage. GitHub OAuth can wait, but leave the code so we can add it later. This has to work in production, not only on localhost.”

saymd writes

## Objective Build Google OAuth login for the Next.js app using the App Router, with room for GitHub OAuth later. ## Context Secure user authentication for saymd via Google OAuth. Session should be cookie-based after a successful Google sign-in. ## Instructions - Implement Google OAuth login with the Next.js App Router - Add rate limiting on the auth callback - Issue a session cookie after successful Google sign-in - Structure the auth layer so GitHub OAuth can be added later ## Constraints - No password storage - Do not store refresh tokens in localStorage ## Acceptance criteria - [ ] User can sign in with Google OAuth - [ ] User receives a session cookie - [ ] Google login works in production - [ ] GitHub OAuth can be added later without a rewrite

--file note.m4a · existing recording, default template

You drop

A voice memo on disk: note.m4a — “Add a dark mode toggle on the settings page, right next to the language picker. Remember the choice in localStorage so it survives a refresh. Don’t touch the marketing site at all. Prefer CSS variables instead of pulling in another UI library, and keep the light theme as the default until the user flips it.”

saymd writes

## Objective Add a dark mode toggle to the settings page. ## Context The control should sit next to the language picker. Light theme remains the default until the user opts into dark mode. ## Instructions - Add a dark mode toggle next to the language picker - Persist the choice in localStorage across reloads - Apply the theme with CSS variables (no new UI library) - Keep the marketing site unchanged ## Constraints - Do not change the marketing site - Prefer CSS variables over a new UI library - Light theme is the default

--template bug · from the mic

You say

“Bug report: on Safari in private browsing, Continue with Google on the login page does nothing useful — you get a blank page and the console shows a cookie error. Chrome is fine. Reproduce by opening the login page in Safari private mode and clicking Continue with Google. Expected is a redirect to Google. It must work in Safari private browsing, not only normal windows.”

saymd writes

## Objective Fix login button not working on Safari. ## Context Failure only appears in Safari private browsing. Chrome (and Safari non-private) work. Console reports a cookie-related error after Continue with Google. ## Steps to reproduce - Open the login page in Safari private browsing - Click Continue with Google - Observe a blank page and a cookie error in the console ## Expected behavior Redirect to Google for OAuth. ## Actual behavior Blank page; cookie error in the console. ## Constraints - Must work in Safari private browsing mode

Why not just use built-in voice?

Built-in voice is chat input inside one tool. saymd is the agent-independent layer: a structured prompt file you keep in your repo and reuse anywhere.

	Built-in voice	saymd
Output	Chat bubble or inline text in one IDE	`.ai/prompt.md` in your project
Structure	Raw transcript	Objective, Context, Instructions, Constraints
Reuse	Stays in that tool's session	Git-tracked, shareable, any agent can `@` it
Templates	—	Feature, bug, and plan specs
Evolve the spec	—	Pro: `--continue`, `--review`, cross-language `--out`

Typical workflow: saymd → commit .ai/prompt.md → @.ai/prompt.md in Cursor, Claude Code, Codex, Copilot, Antigravity CLI, or OpenCode.

Voice-to-prompt, not another dictation app.

saymd creates a durable prompt file you can reuse across agents, repos, and workflows.

Structured Markdown

Turn speech into Objective, Context, Instructions, and Constraints instead of a raw transcript.

Mic or audio file

Dictate live, or skip the mic entirely: saymd --file idea.m4a (mp3, wav, and more). Same structured output either way.

Agent-independent

Use the same prompt file with Cursor, Codex, Claude Code, Copilot, Antigravity CLI, or any tool that can read Markdown.

BYOK by design

Bring Your Own Key — Gemini, OpenAI, Deepgram or ElevenLabs on your account. saymd never sells transcription minutes. Keys stay in ~/.saymd/config.json.

85+ languages

Speak naturally with automatic detection. Free keeps output in the same language.

Cross-language Pro

Think in Hungarian. Ship English specs. Use --out en when your agent or team works in another language.

Simple pricing.

The Free CLI is MIT open source. Pro is a separate commercial package — continue, review, and cross-language output when the prompt becomes part of your daily workflow. Full details on the pricing page.

Free

Open-source MIT CLI

€0

Mic or existing audio via --file
85+ languages with auto-detect
Structured Markdown output
Feature, bug, and plan templates

BYOK voice models

Gemini gemini-3.5-transcribe recommended
OpenAI gpt-4o-transcribe
Deepgram nova-3
ElevenLabs scribe_v2

Get Free on GitHub

Pro

For daily developer workflows

Best value

€39 / year

or €5/month

Everything in Free, plus:
--continue — merge new speech into the same spec
--review — check gaps before you ship
--out en — cross-language output
.saymd/vocab.txt — project vocabulary

Get Pro — €39/year Prefer monthly? €5/month

Manage card, invoices, or cancel in Stripe: Manage subscription. Lost an activation code? Email support@saymd.app.

Feature	Free	Pro
Mic + audio file input	✓	✓
Structured Markdown	✓	✓
Feature / bug / plan templates	✓	✓
85+ input languages	✓	✓
BYOK models `gemini-3.5-transcribe` · `gpt-4o-transcribe` · `nova-3` · `scribe_v2`	✓	✓
BYOK — Bring Your Own Key	✓	✓
Continue existing prompt	—	✓
Prompt review	—	✓
Cross-language output	—	✓
Project vocabulary	—	✓

After purchase.

Your activation code appears on the success page after checkout. Activate it once from the terminal.

activate saymd Pro

saymd activate <activation-code>

Frequently asked questions.

Everything you need to know before installing or upgrading.

What is saymd?

saymd is a terminal CLI that turns microphone recordings or audio files into structured Markdown prompts for AI coding agents. The goal is not just transcription — it is creating a reusable prompt artifact.

Why not just use built-in voice in Cursor, Claude Code, or Copilot?

Built-in voice is great for quick chat — but the transcript usually stays inside that session. saymd writes a structured Markdown file (Objective, Context, Instructions, Constraints) to your repo. You can version it, share it with your team, and reuse it with any agent — not just the one you spoke into.

What does BYOK mean?

BYOK means Bring Your Own Key. You configure your own API keys for Google Gemini, OpenAI, Deepgram, or ElevenLabs — saymd does not sell transcription credits or host speech-to-text. Keys are stored locally in ~/.saymd/config.json. Audio goes to the provider you chose, not to saymd.app.

Which STT providers are supported?

Free includes all four: Gemini (gemini-3.5-transcribe, recommended), OpenAI (gpt-4o-transcribe), Deepgram (nova-3 multilingual), and ElevenLabs (scribe_v2). Run saymd setup to pick one and paste your key.

Which languages are supported?

saymd supports more than 85 input languages with automatic detection. Free output stays in the detected language. Pro adds cross-language output, such as speaking Hungarian and generating an English specification.

Which AI agents does it work with?

Any agent or coding tool that can read a Markdown file can use a saymd prompt. Common examples include Cursor, Codex, Claude Code, GitHub Copilot, Antigravity CLI, and OpenCode.

Does saymd store my audio?

The CLI does not send audio to saymd.app. Audio is sent to the STT provider you chose (Gemini, OpenAI, Deepgram, or ElevenLabs) with your own API key. Local temporary files are kept only until the Markdown write succeeds, then deleted. See the Privacy Policy.

What is included in Free versus Pro?

Free includes voice/audio-file input, 85+ languages, structured Markdown, feature/bug/plan templates, and all four BYOK STT providers. Pro adds --continue, --review, cross-language output, and project vocabulary. There is no free Pro trial — upgrade when those features become part of your workflow.

Can I cancel anytime?

Yes. Use Manage subscription to open Stripe — invoices, card updates, and cancellation live there. Email support@saymd.app if you need help.

Speak the idea. Ship the spec.

Start free, or unlock the workflow built for prompts that keep evolving.

Get Pro — €39/year