Jarvis (MVP)

Personal planning assistant: dashboard, goals, tasks, reflections, typed chat with an OpenAI-compatible LLM, and approval-only writes from conversations.

Stack

• Next.js (App Router) + TypeScript + Tailwind CSS v4
• Prisma ORM + SQLite
• Server Actions for CRUD and suggestion approvals
• API routes: POST /api/chat, POST /api/recommend

Prerequisites

• Node.js 20+ recommended
• An API key for any OpenAI-compatible HTTP API (/v1/chat/completions)

Setup

1. Install dependencies

   npm install

2. Environment

   cp .env.example .env

   Edit .env:

   • DATABASE_URL — default file:./dev.db (relative to prisma/)
   • OPENAI_API_KEY — required for chat and “What should I work on?”
   • OPENAI_BASE_URL — optional; default https://api.openai.com/v1
   • OPENAI_MODEL — optional; default gpt-4o-mini

3. Database

   npx prisma migrate dev
   npx prisma db seed

4. Run

   npm run dev

   Open http://localhost:3000.

Scripts

Command	Purpose
npm run dev	Dev server
npm run build / npm start	Production build / start
npm run db:migrate	Create/apply migrations (prisma migrate dev)
npm run db:seed	Run seed
npm run db:generate	Regenerate Prisma Client

Architecture (modular)

• src/lib/db/ — Prisma helpers (goals, tasks, reflections, suggestions, dashboard snapshot, chat persistence)
• src/lib/memory/build-context.ts — Builds bounded MEMORY_CONTEXT for the LLM
• src/lib/llm/ — HTTP client, prompts, JSON parsing, applying approved suggestions
• src/lib/jarvis/ — Chat and recommendation orchestration (UI-agnostic; easy to call from voice later)
• src/app/api/chat / recommend — Thin HTTP layer with error handling

Chat uses recent messages from SQLite plus current structured memory, not the full transcript. Jarvis returns JSON (message + suggested_updates). Each update is stored as SuggestedUpdate with pending until you approve it on the dashboard or chat page.

Seed data

Includes sample long-term/short-term goals, three tasks, one reflection, and primary focus text matching the Jarvis MVP brief.

Notes

• No auth in this MVP — run locally or behind a trusted network.
• Some OpenAI-compatible servers may not support response_format: { type: "json_object" }; if chat fails, try OpenAI or a compatible proxy, or adjust src/lib/llm/client.ts.
• Local SQLite file: prisma/dev.db (gitignored).