I BUILT an AI Food Tracking App Using Just ONE TOOL (OpenClaw)

The creator built a personal AI nutritionist in OpenClaw that logs meals from a phone photo into a local SQLite DB, beats generic apps for a fatty liver + keto case, but still needs manual correction on portion math.

Why he skipped existing apps

He calls standard nutrition apps "absolute garbage." His constraints don't fit a template: diagnosed fatty liver, keto diet, Asian grocery packaging in Chinese, and a 6-month doctor check-in deadline. No app on the market covers that combo, so he turned to OpenClaw and wrote a user story: lose weight, accept food photos, track macros, build a database of what he generally eats.

The stack and data model

OpenClaw (called "Maximoff" / "her" throughout) runs a local SQLite database. Two columns matter: a photo column so mis-logged items can be audited, and a separation between AI-entered nutrition facts vs user-entered portions — the creator was explicit that the food database values should not overwrite what he actually ate. He also added a persistent "skill" for DB access so the agent doesn't "forget" between sessions (his phrasing for context-window loss).

What worked

A photo of pine nuts, a Coke Zero, a protein shake, and Chinese-language beetroot powder sachets were all parsed correctly. The agent handles mixed-language labels and pulls macros from its own internal food database, then writes a daily log. He targets under 50 g of carbs/day for keto and gets a breakdown he can read in chat.

What broke

The beetroot powder entry was logged as 56 g of carbs because the agent read the per-sachet carbs (10 sachets) as his actual intake. One correction message fixed it. He treats this as expected — the workflow is: photo → log → spot error → type a fix.

Next step

Add a daily weight chart so the agent can correlate weight against macros, then build a custom dashboard instead of relying on chat output.