Calorie AI

Native Android and Dockerized web calorie tracker with meal images, AI nutrition estimates, server-backed plans, and admin-configurable AI models.

Features

• Add meals by description, portion estimate, and uploaded image.
• Analyze meals through OpenAI-compatible chat completions.
• Uses a vision model for food image calorie and portion estimates.
• Uses an advice model to normalize calories, protein, carbs, fat, fruit servings, vegetable servings, food groups, and notes.
• Stores meal entries, settings, plans, and synced activity in PostgreSQL.
• Imports Garmin Connect and gym app activity through Android Health Connect.
• Shows daily totals plus charts for macros, 7-day calories, and fruit/vegetable intake.
• Admin settings control API base URL, API key, image model, and nutrition/advice model. The two model fields can contain the same model name.

Android Build

Open this folder in Android Studio and run the app configuration, or build with Gradle if available:

gradle :app:assembleDebug

Forgejo Actions builds the debug APK on every push to main and uploads it as the calorie-ai-debug-apk artifact.

Tagged versions also build an installable APK named calorie-ai-vX.Y.Z.apk and attach it to a Forgejo release through the Android Release workflow.

Install With Obtainium

Use the Forgejo releases page as the source:

https://git.danvics.com/danvics/calorie-ai-android/releases

If Obtainium asks for a direct APK URL, use the latest release asset URL pattern:

https://git.danvics.com/danvics/calorie-ai-android/releases/download/v0.1.1/calorie-ai-v0.1.1.apk

Recommended Obtainium setup:

• App source: HTML or Gitea/Forgejo if your Obtainium version offers it.
• URL: https://git.danvics.com/danvics/calorie-ai-android/releases
• APK link filter: calorie-ai-.*\.apk
• Version extraction: release tag, for example v0.1.1.

The APK is a debug build, so Android may require allowing installs from Obtainium and accepting the debug signing certificate.

Garmin And Gym Activity

Calorie AI imports external activity through Android Health Connect:

• Garmin: enable Health Connect sync in Garmin Connect, then open Calorie AI on Android and use Settings -> Sync Health Connect.
• Gym machines: use the machine's companion app or gym app to write workouts to Health Connect, then sync from Calorie AI.
• Imported data includes daily steps, active calories, and exercise sessions when Health Connect exposes them.

Direct Garmin Health API sync requires Garmin partner approval and API credentials, so the supported path is Health Connect first.

Web Frontend

The web frontend lives in web/. It serves an authenticated browser UI and a tiny Node proxy at /api/chat so OpenAI-compatible endpoints are called server-side instead of directly from the browser.

Run it with Docker:

cd web
docker compose up -d --build

Then open:

http://127.0.0.1:8095

The Docker web stack includes PostgreSQL. Runtime data lives under web/data/postgres, and legacy JSON/SQLite data in web/data is migrated on startup.

The Docker web server creates credentials on first boot if CALORIE_AI_WEB_PASSWORD and CALORIE_AI_SESSION_SECRET are not supplied. To pin credentials, copy web/.env.example to web/.env and set strong values before starting Docker Compose.

AI Endpoint

The app expects an OpenAI-compatible endpoint at:

{API base URL}/chat/completions

Example base URLs:

• https://api.openai.com/v1
• An emulator host-loopback URL ending in :11434/v1 for a local OpenAI-compatible service

The web and Android clients use the same authenticated server APIs for entries, settings, plans, models, and synced activity.