Vibe Cooking

An AI-powered cooking assistant — plan, tweak, and cook recipes through a relaxed chat interface.

---

Overview

Vibe Cooking pairs a recipe library with a chat assistant. You build and refine recipes by talking to the assistant; it can suggest recipes, save them, and edit them on your behalf. Every change — yours or the assistant's — is versioned, so the assistant can adapt a recipe freely without ever losing the original.

Core Concepts (see prisma/schema.prisma)

Recipes

A Recipe is a stable identity plus lifecycle state (archivedAt). Its content — name, ingredients, instructions, servings, duration, difficulty — lives in append-only RecipeRevision snapshots.

• The current recipe is always the latest revision (max(revision)).
• Editing appends a new immutable snapshot rather than mutating the previous one, so full history is preserved.
• Each revision records its Author (USER or ASSISTANT), so assistant edits can be distinguished from yours — and reverted by copying an earlier revision forward.
• Deletes are soft (archivedAt); history is never destroyed.

User Profile

A free-text profile (dietary needs, tastes, context) that informs the assistant's suggestions. Like recipes, it's stored as append-only UserProfileRevision snapshots (latest = current), attributed via Author.

Assistant

A single chat endpoint parameterized by an app context:

• mainAssistant — the recipe library: browse, get suggestions, save recipes, edit your profile.
• recipeView — a specific recipe: ask questions or request edits in place.

The assistant works through structured tools (recipe CRUD, profile updates, web search, recipe-suggestion rendering). Versioning is invisible to the model — it just creates/updates recipes, and the system records authorship and revisions underneath.

Architecture

A layered architecture keeps concerns separate:

1. Data Access (lib/data-access/) — direct database operations via Prisma.
2. Actions (lib/actions/) — business logic, authentication, user context.
3. API Routes (app/api/) — transport, with Zod validation.
4. Client (lib/api/client.ts) — typed API calls.
5. Hooks (lib/api/hooks/) — data fetching and caching with React Query.
6. Components — the chat-based UI.

AI tools (lib/ai/tools/) are split across definitions (schemas), execution (server logic), renderer (client UI), effects (cache invalidation), and tools (wiring).

Domain vs persistence types: the app speaks in clean domain types (e.g. Recipe in lib/types.ts = identity + current content), while the Prisma row is aliased (RecipeRow) where raw rows are handled. See CLAUDE.md for the convention.

Tech Stack

• Next.js (App Router) + TypeScript
• PostgreSQL with Prisma ORM
• WorkOS AuthKit for authentication
• React Query for state management
• Vercel AI SDK for the assistant
• Radix UI + Tailwind CSS

Getting Started

# 1. Install
npm install

# 2. Configure environment (.env)
#    DATABASE_URL  — pooled connection used at runtime
#    DIRECT_URL    — direct connection used by the Prisma CLI / migrations
#    plus WorkOS AuthKit + AI provider keys

# 3. Apply migrations
npx prisma migrate deploy

# 4. Run
npm run dev          # http://localhost:3000

Local database

For a throwaway local Postgres (migrations read DIRECT_URL, runtime reads DATABASE_URL):

createdb vibe_cooking_dev
DIRECT_URL="postgresql://<user>@localhost:5432/vibe_cooking_dev" npx prisma migrate deploy
# point the dev server at it via .env.local: DATABASE_URL="postgresql://<user>@localhost:5432/vibe_cooking_dev"

Run npx prisma generate after any schema.prisma change (including git stash/checkout) so the client matches.

Project Status

Under active development. Core functionality is stable; UX and features are evolving.