Reed

Reed is a memory-first AI mentor for career decisions, applications, and follow-through. The demo is designed around one core behavior: Reed opens a conversation with useful context before the user types, based on prior goals, patterns, and commitments.

Live app: https://ask-reed.vercel.app

What It Demonstrates

• Proactive session opening based on stored commitments and behavioral signals.
• A chat interface that keeps long-running career context visible without replaying raw transcripts into the UI.
• Resume and document attachments, with extracted text stored separately from the visible message content.
• Press-to-record voice input that transcribes speech into the editable message box without auto-sending.
• Past conversations in a read-only transcript viewer, including attachment chips without exposing raw attachment text.
• Supabase-backed authentication and row-level security.

Tech Stack

• Next.js 16 App Router
• React 19 and TypeScript
• Vercel AI SDK with Anthropic Claude
• Supabase for auth, Postgres, and row-level security
• Tailwind CSS 4
• Groq Whisper for speech-to-text transcription
• pdf-parse for text-based PDF extraction

Project Structure

app/
  api/chat/              streamed chat route and memory persistence
  api/extract-pdf/       PDF text extraction endpoint
  api/session-context/   current memory snapshot endpoint
  api/stt/               voice transcription endpoint
  auth/callback/         Supabase auth callback
  chat/                  authenticated chat page
  login/                 auth page
components/
  login-form.tsx         email/password and Google auth
  reed-app.tsx           main chat experience
  session-memory-peek.tsx memory sidebar and transcript viewer
lib/
  chat/                  model constants and system prompt
  coaching-logic/        session context and proactive nudge logic
  memory/                structured memory extraction
  supabase/              browser, server, and admin clients
supabase/migrations/     database schema

Environment Variables

Create .env.local with:

NEXT_PUBLIC_SUPABASE_URL=
NEXT_PUBLIC_SUPABASE_ANON_KEY=
SUPABASE_SERVICE_ROLE_KEY=
ANTHROPIC_API_KEY=
GROQ_API_KEY=

SUPABASE_SERVICE_ROLE_KEY is server-only. Do not expose it in the browser or prefix it with NEXT_PUBLIC_. ANTHROPIC_API_KEY powers chat responses, and GROQ_API_KEY powers voice transcription.

Supabase Setup

Apply the migrations in order:

supabase/migrations/001_initial_schema.sql
supabase/migrations/002_message_attachments.sql

The second migration adds:

attachment_filename text
attachment_text text

to public.messages. These columns are required for attachment persistence and follow-up questions over uploaded content.

For auth, configure:

Site URL: https://ask-reed.vercel.app
Redirect URLs:
  https://ask-reed.vercel.app/auth/callback
  http://localhost:3000/auth/callback

Local Development

Install dependencies:

npm install

Start the dev server:

npm run dev

Open:

http://localhost:3000

Quality Checks

Run before shipping:

npm run type-check
npm run lint
npm run build

Core Demo Flow

1. Sign in with email/password or Google.
2. Reed should open with a proactive nudge about the unresolved product-adjacent role commitment.
3. Ask a real career or resume question.
4. Use the mic button once and confirm the transcript lands in the editable input without sending.
5. Upload a text-based resume file.
6. Ask about the uploaded content.
7. Send an unrelated message.
8. Ask about the uploaded content again and confirm Reed still uses the attachment context.
9. Open Past Conversations and verify the transcript is read-only, ordered, and renders attachment chips instead of raw attachment text.

Deployment

The app is deployed on Vercel. After pushing to main, Vercel redeploys the production app at:

https://ask-reed.vercel.app