The tree that knows what you eat.
Say what you ate. The AI parses it, tracks your macros, learns your patterns, and tells you what to eat next. One command. The tree does the rest.
One LLM call parses the food into macros. Cascade signals route protein, carbs, and fats to their tracking nodes. The daily picture updates instantly.
The AI reads your remaining macros, your meal history for variety, and your fitness data through channels. Specific recommendations, not generic advice.
Weekly averages, hit rates, and meal slot patterns. The AI finds what's working and what isn't. History node archives 90 days of daily summaries.
No external database. No calorie counting app. The tree structure holds your macros, your meal patterns, your profile, and your history. The AI reads the structure to know everything.
A channel between Food and Fitness carries data both ways. Neither extension imports the other. The tree connected them through structure.
The fitness AI sees your calories. The food AI sees your workout. Both give better advice.
It goes deeper than you think.
Log parses natural language food input into structured macros. One LLM call. Review analyzes patterns. Reads weekly averages, hit rates, meal slot history, and fitness data. Advises forward and backward. Coach sets up goals. Asks about calorie targets, macro splits, and restrictions.
When you log food, channels carry the parsed macros to Protein, Carbs, and Fats nodes. Each node increments its daily total atomically. No read-modify-write. No race conditions. The Daily node reads all siblings through enrichContext. One structure, zero extra queries.
At midnight (synced to the breath extension or a fallback timer), the day's totals archive to History as a note. Weekly averages and hit rates update on each macro node. The AI sees "protein hit rate: 43% this week" without reading 90 history notes.
Navigate to your Food tree and type "be". The AI asks what you've eaten. You answer. It logs, routes, and tells you where you stand. No menus. No calorie lookups. Just a conversation with a tree that counts for you.
Nothing is hardcoded. The scaffold creates Protein, Carbs, and Fats as defaults. After that, the system treats them the same as any node you create yourself.
Log, Daily, Meals, Profile, History. These define the shape. Only Log is required. Everything else is optional.
Everything else. Protein, Carbs, Fats, or whatever you add. Sugar, Fiber, Sodium, Water. The tree tracks whatever exists.
The LLM parser prompt is built at runtime from the tree. If your tree tracks protein, carbs, fats, sugar, and fiber, the prompt says exactly that. Delete protein and add sodium, the prompt changes. The parser outputs the right fields. Cascade signals route to the right nodes. Daily reset archives the right totals. The dashboard renders a progress bar for every metric node. No config file. No code change. Just tree structure.
Create a plain node under your food tree called Sugar. Next time you log food, the AI notices it and asks: "Sugar isn't tracked yet. Want me to add it? What's your daily goal?" One conversation. The node gets its metadata, joins the pipeline, and starts tracking.
Structural nodes are protected across every extension. Any node with a role in any extension's metadata namespace is guarded from deletion. The system names the extension and role. --force bypasses it when you mean it.
Plant a land. Create a Food tree. Say what you ate. The tree takes it from there.