# How to Vibe Code Your Own Supernormal (and Stop Paying for It)

> AI agent platform to turn meetings into completed client work

- Site: https://supernormal.com
- Category: AI Meeting Assistant & Agent Suite
- Verdict: **Serious undertaking** (35/100 vibecodeable)
- Estimated effort: 4-6 weeks of part-time work

## Verdict

Build a personal web-based clone with file uploads and LLM template generation, but skip native system audio capture unless you enjoy debugging low-level OS kernel extensions.

Supernormal combines a standard web application dashboard with a brutally hard native desktop utility that hooks into local audio cards without meeting bots. While you can easily vibecode the Next.js web application, rich text notes, project folders, and LLM prompt templates using Claude Code and Cursor, reproducing the native macOS/Windows system-audio capture driver and real-time speaker separation requires deep platform-specific low-level systems programming that AI agents will struggle to scaffold reliably.

### What you can't replicate

- Native kernel-level audio capture drivers operating across macOS and Windows without bot injection
- Enterprise trust, SOC 2 Type II compliance, and HIPAA certifications
- The massive active user base and institutional backing network

## What it does

Supernormal provides bot-free desktop recording for meetings and leverages AI agents to transform unstructured meeting transcripts and context into rich client deliverables such as presentations, documents, spreadsheets, and MCP context feeds.

### Core features

- Cross-platform desktop native wrapper (macOS menu bar / Windows taskbar) for local audio capture
- Bot-free audio recording with mic switching and dynamic audio device routing
- Speech-to-text transcription with speaker diarization
- Post-meeting command center with rich-text editor (Notion-style)
- Multi-step AI agent task orchestration for structured artifact generation (presentations, docs, spreadsheets, images)
- Project folders organizing multiple meetings and documents into shared context
- Model Context Protocol (MCP) server integration to feed meeting context into external AI tools (Claude, Cursor)
- Shortcuts and prompt templates for instant summaries and deliverables

## The business

### Pricing

- Free: $0 — For individual evaluation and limited agent generation.
- Team: $0 base tier structure — Team collaboration with pooled credits and advanced rollover.
- Business: Custom — Enterprise-grade governance and security controls.

### Funding

$12.9M raised.
- Seed Round (Dec 2020)
- Seed Extension (Jan 2023)
Investors: EQT Ventures, Balderton Capital, byFounders, Acequia Capital

Founded 2020.
Team size: 20-35.

## The hard parts

- Capturing system audio and microphone feeds natively across macOS (Apple Silicon kernel/audio extension permissions) and Windows 11 without injecting a virtual calendar bot
- Orchestrating multi-step LLM generations that reliably build complex structured assets (like multi-page presentation decks or spreadsheets) without hallucinations
- Real-time speech transcription pipelines with clean speaker diarization and audio buffering during device switches

## How to vibe code Supernormal

### Prerequisites

- Node.js (free): Runtime for running Next.js web application and build scripts.
- GitHub (free): Version control and repository hosting for your codebase.

### Recommended AI tools

- Claude Code: Terminal agent to scaffold the full Next.js stack, database schema, and LLM processing pipelines end-to-end.
- Cursor: Code editor for fine-tuning frontend components, rich-text markdown editors, and prompt templates.

### Stack

- Frontend: Next.js with Tailwind CSS and shadcn/ui components
- Backend: Next.js API routes / Server Actions
- Database: Turso (libSQL/SQLite at the edge)
- Auth: better-auth
- Payments: none
- Other: Vercel AI SDK, OpenAI API (Whisper STT + GPT), Anthropic API (Claude Sonnet for agentic generation)

### Hosting

- Vercel (Hosting the Next.js web application and serverless API endpoints): $0-20/mo
- Turso (Serverless edge database for storing meeting transcripts, notes, and project metadata): $0/mo

### Build guide

1. **Project Scaffolding and Database Schema** — Initialize the Next.js project with Tailwind CSS, shadcn/ui, Turso connection via Drizzle ORM, and better-auth for single-user management.

```
Scrape together a new Next.js project using Tailwind CSS and TypeScript. Set up Turso database connectivity using Drizzle ORM with tables for 'users', 'projects', 'meetings' (storing title, transcript, raw audio URL, status, and summary), and 'deliverables' (storing type, content, and metadata). Configure better-auth for local user authentication. Ensure all configuration files are complete and clean.
```

2. **Audio Upload and Transcription Pipeline** — Build an audio file upload interface in the web app and connect it to OpenAI's Whisper API for batch transcription and speaker diarization.

```
Create a meeting capture upload page in Next.js where users can drag and drop audio recordings (.mp3, .wav, .m4a). Implement a server action that uploads the file to object storage and calls the OpenAI Whisper API to get the transcript with timestamps and speaker tags. Save the resulting transcript JSON directly to the Turso meetings table.
```

3. **Rich Text Meeting Notes and Project Folders** — Develop the post-meeting dashboard featuring a Notion-style rich text editor for notes and project grouping drawers.

```
Build a post-meeting workspace page using Tailwind CSS with a split-pane layout. The left side features a rich-text markdown editor for meeting notes that auto-saves changes to the backend. The right side contains a collapsible AI chat and shortcuts panel. Add a Projects management view that allows grouping multiple meetings into a shared project container.
```

4. **Agentic Deliverable Generation** — Implement Vercel AI SDK integrations to let AI agents generate structured deliverables (presentations, briefs, one-pagers) from meeting context.

```
Integrate the Vercel AI SDK with Anthropic Claude Sonnet to power an agentic generator. Create template actions that ingest a meeting's transcript and project notes to output structured JSON representations of slide decks, creative briefs, or budgets. Build a UI renderer that displays these generated artifacts inside the workspace and allows exporting them.
```

5. **Model Context Protocol (MCP) Server Integration** — Expose an MCP endpoint or local server config so external developer tools like Cursor and Claude Desktop can read meeting history and context.

```
Implement an MCP (Model Context Protocol) server route within the Next.js backend that exposes tools to query past meeting notes, transcripts, and project summaries by keyword or project ID. Include setup instructions and a configuration payload snippet for connecting Claude Desktop or Cursor to your local MCP server.
```

### Cost vs paying

**Starting costs (one-time):**

- Domain name registration: $12 one-time
- OpenAI/Anthropic API starter credits: $10 one-time
- Total: ~$22 one-time

**Ongoing costs (monthly):**

- Vercel Hobby / Turso / Cloudflare R2: $0/mo
- OpenAI Whisper & Claude API usage (personal volume): ~$5/mo
- Total: ~$5/mo

- Paying for the SaaS instead: ~$18-50/mo
- Build time: 40-60 hours
- AI tool credits: $20 (Claude Pro or Cursor Pro)
- Break-even: N/A (Personal clone)

## Sources

- [Supernormal Official Website](https://supernormal.com)
- [Supernormal Pricing](https://supernormal.com/pricing)
- [Supernormal Product Updates & Changelog](https://supernormal.com/blog)
- [Supernormal About Page](https://supernormal.com/about)