food-water-log

Food & Water Log

What It Does

A voice-powered food and water intake tracker that persists across sessions. Log meals and water hands-free, review today's intake, get weekly summaries — all saved to persistent storage so your data survives between conversations. The LLM classifies natural speech so you can say things like "I had eggs for breakfast" or "drank 2 glasses of water" naturally.

Why this needs to be an Ability: The LLM can't persist data across sessions on its own. This Ability uses the File Storage API to save and recall your food and water log over time — something the Agent alone cannot do.

Suggested Trigger Words

• "food log"
• "water log"
• "log my food"
• "log my water"
• "what did I eat"
• "track my food"
• "track my water"
• "food tracker"
• "meal log"
• "hydration"

Setup

No API keys or external services required. Uses only the built-in LLM and persistent file storage.

How It Works

1. User triggers the ability with a hotword (e.g., "food log")
2. Ability loads persistent data from oh_food_water_log.json
3. Returning users see today's summary; first-time users get a welcome prompt
4. User speaks naturally — the LLM classifies intent and extracts data
5. Entries are saved to persistent storage immediately
6. User can log more, read back entries, get summaries, or say "done" to exit

Supported Commands (via natural language)

Intent	Example Phrases
Log food	"I had eggs and toast", "pizza for lunch", "coffee and a bagel"
Log water	"drank a glass of water", "2 glasses of water", "had some water"
Read today's food	"what did I eat today", "show my meals"
Read today's water	"how much water did I drink", "water count"
Read full log	"read my log", "what did I have today"
Weekly summary	"weekly summary", "how was my week"
Clear today	"clear today" (with voice confirmation)

Example Conversation

User: "food log" AI: "Food and water log is open. Tell me what you ate or drank, or ask what you've had today." User: "I had eggs and toast for breakfast" AI: "Logged eggs and toast for breakfast. That's 1 meal today." User: "drank 2 glasses of water" AI: "Logged 2 glasses of water. Total today: 2 glasses." User: "coffee and a muffin" AI: "Logged coffee and a muffin. That's 2 meals today." User: "what did I eat today" AI: "Today's food: eggs and toast for breakfast, coffee and a muffin. 2 entries total." User: "how much water" AI: "You've had 2 glasses of water today." User: "done" AI: "Log closed. Stay hydrated!"

Next session:

User: "food log" AI: "Welcome back! Today you've logged 2 meals and 2 glasses of water. What would you like to add?"

Key SDK Functions Used

SDK Method	Purpose
speak()	Voice output to user
user_response()	Listen for voice input
text_to_text_response()	LLM intent classification with system prompt
run_confirmation_loop()	Yes/no confirmation before clearing entries
check_if_file_exists()	First-run detection
read_file()	Load persistent log data
write_file()	Save log data (with delete-first pattern for JSON)
delete_file()	Clear old JSON before writing updated data
resume_normal_flow()	Return to Agent (in try/finally)
editor_logging_handler	All logging (no print statements)

Architecture

• LLM as intent router — classifies natural speech into structured JSON intents
• Persistent file storage — oh_food_water_log.json survives across sessions
• Delete + write pattern — prevents JSON corruption from append behavior
• Namespaced file — prefixed with oh_ to avoid collisions with other abilities
• First-run vs returning user — greeting adapts based on existing data
• Idle detection — 2 consecutive empty inputs triggers graceful exit offer
• Voice confirmation — clearing entries requires spoken yes/no
• try/finally — guarantees resume_normal_flow() on every exit path