📖 Playbook

AI-powered SOP knowledge base. Ask questions about your company procedures and get step-by-step answers from your documentation.

Stack

• Next.js 15 (App Router)
• Supabase (Postgres + pgvector)
• OpenAI (embeddings + chat via Vercel AI SDK)
• shadcn/ui + Tailwind CSS

Setup

1. Supabase

Create a Supabase project, then run the migration:

# Via Supabase dashboard: SQL Editor → paste contents of supabase/migrations/001_initial_schema.sql
# Or via CLI:
supabase db push

Note the id of the inserted "Demo Company" client — you'll need it for DEFAULT_CLIENT_ID.

2. Environment Variables

cp .env.example .env.local

Fill in:

• NEXT_PUBLIC_SUPABASE_URL — from Supabase project settings
• NEXT_PUBLIC_SUPABASE_ANON_KEY — from Supabase project settings
• SUPABASE_SERVICE_ROLE_KEY — from Supabase project settings (service role, not anon)
• OPENAI_API_KEY — from OpenAI dashboard
• DEFAULT_CLIENT_ID — UUID of the demo client row

3. Install & Run

npm install
npm run dev

4. Seed Data

With the dev server running:

npx tsx seed/seed.ts

This creates 3 sample SOPs (Client Onboarding, Invoice Processing, New Employee Setup) with embedded chunks.

5. Use

• / — Chat interface. Ask questions like "How do I onboard a new client?"
• /admin — Manage SOPs and view analytics

Architecture

• SOPs are split into ~400-token chunks at ## heading boundaries
• Each chunk is embedded via OpenAI text-embedding-ada-002
• Chat queries are embedded and matched via pgvector cosine similarity
• Matched chunks are injected as context for GPT-4o-mini to answer
• Query logs track usage, confidence, and feedback for analytics