Log a meal in plain language

Two in the afternoon. You finish your plate, push it aside, and pull out your phone before the post-lunch fog hits. You didn't follow the diet plan exactly — there was a side salad you didn't plan for, and the rice was a generous serving. The question is how to log it without opening a food picker, scrolling through a list, or doing the math on the macros.

What the agent needs to hear

In Verxion Native voice (limited rollout): no workout is needed. Say “Log lunch today: 180 grams of chicken and 200 grams of cooked rice.” The agent may ask for missing details; today’s dictated meal can be recorded without a second confirmation. Creating a reusable saved meal or changing another day’s records requires a spoken confirmation. Check the meal actually saved in your log. The example card is for an MCP client; a Native call reports the result by voice and in the conversation. See the voice guide.

Behind your sentence, the agent is working backwards to four pieces of information: which meal (breakfast, lunch, dinner, snack), what was on the plate, how much of each, and when you ate it. You don’t have to spell every one out. Time of day fills in the meal type most of the time, and “today” is the default unless you say otherwise. But if the meal is unusual for the hour, or you’re logging something from earlier, naming it explicitly removes guesswork: « log this as lunch even though it’s 4 pm » lands cleanly.

For the items themselves, keep the natural names — « chicken », « rice », « salad ». The agent searches your catalog by what you said. A widget-capable MCP client may show the matches in a card; in Native voice, listen to the result and review the saved meal. It doesn’t need a brand, a barcode, or a precise variety unless you have multiple variations and want to disambiguate (« the basmati, not the jasmine »).

Portions and defaults

Give measured portions when you have them: « 180 g chicken, 200 g cooked rice ». If you only know an estimate, say so explicitly. The agent may need to ask about a missing amount or an ambiguous unit; do not treat an assumed portion as a measurement.

If the meal had a repeat structure — your usual breakfast smoothie, the lunchbox you packed twice this week — mention it by name (« my usual post-workout shake ») and the agent pulls the saved version instead of resolving each ingredient one by one. Faster, and consistent across days.

When the match looks wrong

In a widget-capable MCP client, review the matched food on the card. In Native voice, check the meal saved in your log and request a correction when necessary. If « chicken » got resolved to breaded chicken when you had it grilled, correct it in plain language: « the chicken should be grilled, not breaded ». If a quantity is off, say the right one — you don’t need to point at the row: « the rice was 200 g ». You’re not editing fields, you’re talking; the agent finds the line and updates it.

If the agent can’t find a food at all — something new you haven’t logged before — it asks. Describe it (« it was a protein bar, RXBar, peanut butter ») and it gets looked up in the shared catalog or saved as a custom entry for next time.

What you can leave out

You don’t need a diet plan to log a meal. You don’t need to pre-create the meal type, the day, or anything else — the agent creates the log on the spot and attaches everything to it. If you happen to be on a plan and the meal matches a planned slot, it gets tagged that way for adherence. If it doesn’t, off-plan meals count toward your daily totals the same as planned ones; they just don’t move adherence numbers.

What makes the log worth keeping

Three things decide whether this log helps you later: the meal type is right (so daily structure stays clean), the foods are the ones you actually ate (so the macros are real), and the portions are honest — not exact, but honest. The system reads these logs for daily totals, weekly reviews, streaks, and adherence scoring; vague portions are fine, but invented foods or wrong meal types corrupt the signal. Say what you ate in the words you’d use to a friend, review the saved result, and correct any mismatch.