Meeting notes
from a recording

Record a meeting in the browser or upload a recording. Cartesia's ink-whisper model turns the audio into text. An LLM then writes a title, a summary, the decisions and the action items with owners.

Deploy to Vercel View on GitHub

Notes for a 15 second meeting called Pricing sync: a three sentence summary, two decisions, and three open action items with Priya and Marcus named as owners, next to Notes and Transcript tabs and a Copy as Markdown button

Screenshot. The template running locally.

Models: ink-whisper
Languages: 17 in the menu
License: Apache-2.0
Stack: Next.js 16React 19AI SDKIndexedDBshadcn/ui
Cartesia products: Ink

What it does

Meetings are saved in your browser, so there is no database and no login. Each meeting can be copied as Markdown. A separate page collects the open action items from every meeting.

Run it yourself

1
Install
Run the commands below with Node 22.22 or newer. pnpm works too. Fill .env.local from the table below. If a key is missing, the page says which.
npx degit cartesia-ai/cookbook/examples/meeting-notes meeting-notes cd meeting-notes # add CARTESIA_API_KEY, LLM_MODEL and the matching LLM key cp .env.example .env.local npm install npm run dev
Environment variables
CARTESIA_API_KEY
play.cartesia.ai/keys
LLM_MODEL
provider/model, for example openai/gpt-5.4-mini
OPENAI_API_KEY
If LLM_MODEL starts with openai/
ANTHROPIC_API_KEY
If LLM_MODEL starts with anthropic/
AI_GATEWAY_API_KEY
For any other provider. Not needed on Vercel
APP_PASSWORD
Optional locally. Required on Vercel, or none to make the app public
TRUST_PROXY
Optional. 1 behind your own reverse proxy
Record a meeting
Open the local URL the dev server prints and press Start recording. Talk for a few seconds. Then press Stop and write notes. To use an audio file instead, switch to the upload tab above the recorder.

Build it with your coding agent

One prompt that tells your coding agent where to start and what to read first.

Prompt

Claude Code Cursor Codex

Build a meeting notes app with Cartesia speech to text. Start from https://github.com/cartesia-ai/cookbook/tree/main/examples/meeting-notes and follow its AGENTS.md for the commands, files and rules. The server route transcribes the recording with client.stt.transcribe from @cartesia/cartesia-js and model ink-whisper (ink-2 is streaming only and does not work on the batch endpoint). The AI SDK writes the summary, decisions and action items with generateText and a zod schema. The browser posts audio to the route and never sees CARTESIA_API_KEY. If I don't have a Cartesia API key yet, follow https://www.cartesia.ai/agent-quickstart.md to get one.

How it works

RecorderThe browser records at 24 kbps, or you upload a file

ink-whisperCartesia turns the whole recording into text in one request

NotesAn LLM writes the summary, decisions and action items

IndexedDBThe transcript and notes, never the audio

Why it works this way

The transcript is saved first

The page saves the transcript as soon as ink-whisper returns it, then asks for the notes. If the summary fails, the meeting keeps its transcript and shows a Retry button that asks the LLM again.

A finished recording uses the file model

Cartesia has two speech to text models. Ink 2 transcribes live audio as it streams and does not accept a finished file. ink-whisper takes the whole recording in one request, which is what this app sends.

Uploads stop at 4 MB

Vercel rejects request bodies over 4.5 MB. So the app caps uploads at 4 MB, and the recorder stops at 3.6 MB or 20 minutes. On another host you can raise the limit in lib/limits.ts.

Make it yours

Transcribe another language

Pick the language in the Spoken language menu before you record. To add one to the menu, extend the list in lib/limits.ts. The model accepts many more ISO 639-1 codes.

Change what the notes contain

Edit the schema in lib/schema.ts to add or remove fields. Edit lib/prompt.ts to change the instructions the LLM gets.

Store notes on a server

Replace the five async functions in lib/store.ts with calls to your database and keep the hooks.

Join the teams making the switch to Cartesia

Ranked #1

in Speech Arena leaderboard & Speech to Text leaderboard by Artificial Analysis

Questions

Where does the meeting audio go?

The browser sends it to the app's own server, which passes it to Cartesia for transcription and then drops it. The transcript goes to your LLM provider for the notes, and your Cartesia key never leaves the server. Get everyone's consent before you record.

Can I get a live transcript while the meeting is happening?

Not with this app. It sends the whole recording when you stop. For live captions, stream the microphone to Cartesia Ink 2 instead.

Which languages work?

The menu lists 17, from English and Spanish to Japanese and Ukrainian. The LLM writes the notes in the language of the meeting.

What does a meeting cost?

Transcription costs 1 Cartesia credit per 2 seconds of audio, silence included, so a 30 minute meeting is 900 credits. The notes add one LLM call on your own provider key.

Does it know who said what?

No. The transcript has no speaker labels, so an action item gets an owner only when someone says that person's name. The LLM is told not to guess.

Is there a step-by-step guide to this template?

Yes. The recipe walks through the code, the README covers setup, environment variables and deployment, and the batch transcription API reference explains the Cartesia side.