Caltext

Caltext

iMessage calorie tracking assistant powered by AI.

Stack

• Runtime: Bun + Turborepo monorepo
• API: Hono on Nitro (deployed to Vercel, 3 regions)
• iMessage: Chat SDK + Sendblue adapter
• AI: AI SDK v6 + GPT-4.1 (vision + agent)
• Database: Upstash Redis (global, 3 regions)
• Workflows: Vercel Workflow SDK for durable pipelines
• Nutrition: USDA FoodData Central API

Setup

1. Install dependencies

1
bun install

2. Configure environment

1
cp .env.example .env

Fill in the required keys:

3. Run locally

1
bun run dev

4. Deploy to Vercel

1
vercel deploy

Deploys to 3 regions: US East (iad1), London (lhr1), Tokyo (hnd1).

5. Set Sendblue webhook

Point your Sendblue incoming message webhook to:

1
https://your-app.vercel.app/webhooks/sendblue

Project Structure

1
caltext/
2
  apps/
3
    api/                  # Hono API server
4
      src/
5
        index.ts          # Routes + webhook handler
6
        bot.ts            # Chat SDK singleton
7
        router.ts         # Onboarding vs assistant routing
8
      workflows/
9
        handle-message.ts # Main message handler
10
        onboarding.ts     # Multi-step onboarding
11
        reminder-loop.ts  # Daily reminders + summaries
12
  packages/
13
    ai/                   # AI agent + tools
14
    db/                   # Upstash Redis data layer
15
    shared/               # Types, locale, timezone utils

How It Works

1. User texts the Caltext number via iMessage
2. Sendblue forwards the message via webhook
3. New users go through conversational onboarding (name, stats, goal)
4. Returning users interact with the AI assistant
5. Photos are analyzed with GPT-4.1 vision, then grounded in USDA nutrition data
6. Text descriptions are matched against USDA database directly
7. Daily reminders at breakfast/lunch/dinner times (timezone-aware)
8. End-of-day summaries with calorie/macro breakdown
9. Weekly recaps with progress bars and trends