calorie-coach-dashboard

Calorie Coach

A personal calorie coach built end-to-end on One — the agent platform that handles the scheduler, the Telegram channel, the LLM calls, and every integration in one place.

You message a Telegram bot with what you ate. The agent estimates macros, writes a row to your Notion database, and replies in a line. Every Sunday at 8 PM the agent reads your last 7 days, sends a coach letter on Telegram, mails you a designed HTML recap, and waits for a YES/NO reply that updates your goals. The dashboard in this repo is just the read side — six screens of dark mission-control UI rendering whatever the agent has written to Notion.

Why this is built on One

The whole point of this build is what isn't in the repo:

• No backend code. No webhook server, no Lambda, no message queue, no cron service.
• No infrastructure. No Docker, no Kubernetes, no Render/Fly/Railway deploy.
• No proprietary database. Your meals live in your own Notion, not in some vendor lock-in datastore.

One handles all of it natively. The agent owns its scheduler (5 cron rows). The agent owns the Telegram channel (no relay needed). The agent owns the LLM calls (configure once, no SDK plumbing). The agent owns the integrations (Notion, Gmail, Deepgram — all native One connections). The agent's "skills" are three Markdown files pasted into the panel.

Create your own at app.withone.ai/agents.

This dashboard reads from Notion via One's passthrough API using a single ONE_SECRET bearer, so no per-platform credentials live in this app either.

What's in this repo

Six screens

Three skills

These live in skills/ and are documented in detail at skills/README.md.

Setup

The dashboard is just the read side. You still need (a) two Notion databases shaped a specific way, (b) an agent on One with three skills + five schedules + three connections. Walk through these in order:

1. Create the two Notion databases

Inside any Notion page, create two databases with the exact column names below. Names and types matter — the skills query them by name.

Meals (one row per meal)

Targets (one active row holding daily goals)

Add one row to Targets with Active = true and your starting daily goals (e.g. 2200 kcal / 140g protein / 70g fat / 240g carbs). The three Pending columns stay empty — the weekly recap writes them when it has a suggestion, and apply-targets clears them after you reply YES/NO.

Now grab each database's data source ID. Open the database as a full page in Notion, look at the URL: notion.so/<title>-<HEX>?v=... — the hex is the database ID. To get the data source ID (Notion's 2026-03-11 API distinguishes them), use the One CLI: one --agent actions execute notion <retrieve-database-action> <connectionKey> --path-vars '{"databaseId":"<the-hex>"}' and read data_sources[0].id off the response. Or query the database via the dashboard's /api route once ONE_SECRET is set.

2. Configure the dashboard

cp .env.example .env.local
# fill in ONE_SECRET + the two NOTION_*_DATA_SOURCE_ID values
# (NOTION_TARGETS_DATA_SOURCE_ID is optional — falls back to env-var defaults)

npm install
npm run dev
# open http://localhost:3000

3. Set up the One agent

Go to app.withone.ai/agents and create a new agent. Then in the agent's panels:

• Connections — connect Notion, Gmail, Deepgram. Attach all three to the agent.
• Channels — connect a Telegram bot.
• Skills — paste the three files from skills/ into the skills panel. Before pasting, find-and-replace these placeholders in each skill file:
  • <NOTION_MEALS_DATA_SOURCE_ID> → your Meals data source ID from step 1 (in meal-logger.md and weekly-recap.md)
  • <NOTION_TARGETS_DATA_SOURCE_ID> → your Targets data source ID from step 1 (in weekly-recap.md and apply-targets.md)
  • https://calorie-coach-dashboard.vercel.app → your own deployed dashboard URL (in weekly-recap.md, line ~293, inside the EMAIL_TEMPLATE's <a href>). This is the URL the "OPEN MISSION CONTROL →" button in the email points to. If you skip this step, the email will link to the maintainer's deployment instead of yours.
• Schedules — add the five rows listed in skills/README.md.

That's it. Send a meal to your bot on Telegram, and the dashboard will show the row within seconds.

Environment variables

Timezone

All times are formatted in Asia/Kolkata regardless of where the Next.js server runs. See lib/time.ts. Day-key computation (used by aggregateByDay, mealsForDate, todayKey) also uses IST so a meal logged at 01:00 IST is grouped into the correct IST day, not UTC. Fork this and change TZ if you're elsewhere.

Known limitations

• Photos aren't stored. The agent receives photos via Telegram and identifies the food via vision, but the photo itself isn't uploaded anywhere. The meal detail screen shows the photo panel only when photoUrl is set on the row, which it currently never is. To enable this you'd need to add an object-storage connection (S3, Cloudinary) and have the meal-logger skill upload before writing the Notion row.
• The ingredient breakdown is an even split of the total macros across the items the agent recognised — not real per-ingredient data. The Notion Meals schema only stores totals plus a comma-separated Items text field, so the dashboard divides totals by item count and labels the panel ESTIMATED SPLIT. Adding real per-ingredient macros would mean a JSON column on Meals plus updates to meal-logger and the meal-detail screen.

Design credit

UI was exported from the Claude Design handoff bundle and ported to React components. The six-screen mission-control aesthetic and all SVG charts come from that bundle.

License

MIT.