exercise_create
ChatGPTCreate a custom exercise only when the user explicitly wants a movement that is not already available. Repeating a normalized name is retry-safe. Example: {"name":"Landmine Press"}
exercise_progress
ChatGPTGet current personal records and chronological completed-session progress for one exact exercise. Use this for questions about an exercise PR or whether the user is progressing. If the user names an exercise instead of supplying an exercise_ref, call search_exercises first and never guess between distinct or ambiguous matches. Records include heaviest load and estimated one-rep max (Epley, from sets of 1 to 10 reps) and are derived from the whole completed history, as are sessions_total, first_performed_on, set_records (the best load at each rep count from 1 to 12) and record_history (the newest 10 record steps of each metric with the set that set each; record_history_truncated and record_history_total say when older steps were left out). Sessions include working sets, top set, estimated one-rep max, volume in kilograms, and — for duration or distance exercises such as cardio — total working duration in seconds and total distance in meters, so you can compare times, distances, and pace across sessions. Without a date range it returns at most 52 recent sessions; with performed_from or performed_through it returns up to 200 sessions in that range. Check sessions_truncated before treating the series as complete or using its first session as a lifetime baseline. Example: {"exercise_ref":"exr_AbCdEfGh1234"}
find_workouts
ChatGPTFind the authenticated user's GoodSet workout history. Use for a request routed to GoodSet about their last, latest, or recent workout; last chest, back, shoulder, arm, or leg day; last time they performed an exercise; or previous weights, sets, and reps. Do not call this tool when the user explicitly asks to use another app. Omit filters for latest workouts overall. Results contain complete workouts in newest-first order; use training_location_ref when machine context matters, and use the second result for comparisons with the time before. Example: {"area":"chest","limit":2}
food_manage
ChatGPTCreate, revise, or archive a private food from user-supplied label values. Actions: create, revise, archive. A revision replaces the entire composition for future selection; past intake keeps its original values. Use nutrient codes energy_kcal, protein_g, fat_g, carbohydrate_total_g or carbohydrate_available_g as defined by the label, fiber_g, sugars_g, saturated_fat_g, sodium_mg, calcium_mg, iron_mg, potassium_mg, vitamin_c_mg. Omit unknown values. Portion equivalent_amount uses the reference unit. Requires nutrition:read and nutrition:write. Example: {"request":{"action":"create","name":"My yogurt","nutrients":[{"code":"energy_kcal","amount":60},{"code":"protein_g","amount":10}]}}
nutrition_export
ChatGPTExport your nutrition entries, exact nutrient snapshots, referenced food profiles, and all private food revisions, including archived and unused foods, plus body weights, dated target changes, and saved meals. Collect every page using next_cursor until null; deduplicate food profiles by food_version_ref. The versioned JSON is portable. Requires nutrition:read. Example: {"request":{"limit":100}}
nutrition_history
ChatGPTRead food entries and daily nutrition totals for up to 31 days; default to today in the user's time zone. Each day also carries the targets in force with remaining amounts and that day's body weight; the first page also lists saved meals with each item's status, problem, and what logging it now records (logs_as); later pages carry saved_meals null. Page entries with next_cursor. Daily totals cover all entries in the date range, including entries on other pages. Null totals mean incomplete or qualified nutrient data, not zero; a null remaining is partial and remaining_at_most bounds it. Set view to trends (outcome trends) for up to 366 days, default the 90 days ending today: body weight readings (warmup marks those before date_from), an exponentially weighted trend with its state (updating, holding, insufficient) and change over the range; intake per bucket (week by default, day for daily figures) with each nutrient's status (empty, complete, partial) against the targets in force; 7, 30, and 90-day averages over complete days only with the partial and empty days they leave out; and the 30-day average intake beside the trend change. Weekly buckets leave the per-day readings and trend null. Requires nutrition:read. Example: {"request":{"view":"trends","bucket":"week"}}
nutrition_track
ChatGPTWrite the nutrition journal. Food actions: log, correct, remove, repeat, select_food, confirm, unstage, discard. log, log_meal, and repeat stage foods for one day and local hour (consumed_on and hour 0-23; or consumed_at; or meal_slot, which stages breakfast at 8, lunch at 12, snack at 15, and dinner at 19; with none, the current hour today; the summary names the hour) in the same list the GoodSet app shows, and return every staged food in preview_items, the projected day, and a confirmation_token; nothing is recorded yet. Another log for the same hour replaces the foods your earlier log staged there. Foods the user added in GoodSet stay and show added_via app: read them back too. unstage removes one staged food by line_ref, with its revision as expected_version; discard clears the hour's list (name the hour as log does). confirm records the whole list as one entry; if the list changed in GoodSet first, it returns draft_changed with a new token, or draft_cleared, and records nothing. Saved meals: log_meal stages a saved meal named by meal or meal_ref using each food's current label; save_meal saves the most recent entry, an entry_ref, or items under the name in meal (pass meal_ref and expected_version to rename or replace; items restate the whole list, and an item resent exactly as listed is kept as is); delete_meal. Body weight: record_body_weight sets one stated scale reading per day (a new value replaces it; the unit defaults to the user's preference); remove_body_weight. Targets: set_targets sets daily nutrient targets from effective_on (default today); omit an amount to remove that target, set inherit to undo that day's change; unmentioned targets stay. correct and remove return a preview; saved-meal, body-weight, and target actions apply immediately and return their receipt. Grams are the default; oz, ml, fl_oz, cup, tbsp, and tsp work where the food's measures mark them available; portion requires a returned portion_ref. Correct/remove default to the last item of the most recently recorded entry; use returned refs or item_number when targeting another item. Follow next_action and confirm prepared values only when they match the user's instruction. Retry the same confirmation_token after an uncertain commit; never prepare another log to retry it. Requires nutrition:read and nutrition:write. Example: {"request":{"action":"log","hour":12,"items":[{"food_name":"Rice, white, cooked","quantity":180,"unit":"g"}]}}
personalization
ChatGPTShow or change GoodSet interpretation, display, interaction, and notification policy settings. Ordinary workout tools already apply these settings; do not call this before a workout unless the user asks to inspect or change them. Actions: show, change_preferences, change_notification_policy. Example: {"request":{"action":"change_preferences","expected_version":1,"display_weight_unit":"kg"}}
search_exercises
ChatGPTSearch GoodSet exercises by natural name. Use this for discovery, not before every workout set: workout_track resolves ordinary exercise names itself. Results are ranked and may include close-but-different movements from fuzzy matching; an empty list means nothing similar exists. When the best result names a different movement than the user said, confirm with the user before logging against it. Each catalog result carries its family (for example cardio or strength), which distinguishes same-named machines from floor exercises and identifies conditioning work. Equal normalized catalog and custom names are returned separately; never guess. Example: {"query":"bench press","limit":5}
search_foods
ChatGPTSearch GoodSet's locally stored food database and your private label foods by name or barcode. Return food versions, preparation, nutrient basis, portions, loggable units (measures), and sources. Preserve ambiguity between brands and raw/cooked foods. Do not guess portions or calculate nutrients yourself. Requires nutrition:read. Example: {"request":{"query":"Rice, white, cooked","limit":5}}
training_location_find
ChatGPTList the authenticated user's saved GoodSet training locations. This tool does not search real-world places and accepts no place name, address, city, country, or coordinates. Match user-supplied place context against the returned saved names. An empty items list means the user has no saved locations. If no returned name clearly matches, do not delay a workout or require a location. Example: {"limit":20}
training_location_manage
ChatGPTCreate, update, archive, or restore a saved training location. Create only when the user explicitly asks to save a user-provided name. The model-facing contract accepts only that name, an optional user-supplied category, and the reference and version needed for later changes. Address, coordinate, and provider metadata stays outside this tool, and updates preserve metadata maintained by trusted clients. Actions: create, update, archive, restore. Example: {"request":{"action":"create","name":"Basque Fitness","location_kind":"gym"}}
workout_current
ChatGPTShow the authenticated user's current active workout. Use this only when the user asks to see or check the workout they are doing now, or when you need the latest set_number before correct_set. For 'I just started a workout', 'I'm working out now', or a start/resume request, use workout_track action=open instead, even if a workout is already active. The last entry in workout.current_exercise.sets carries it. workout_track results already include the updated workout context, so do not call this tool after a workout_track change. This tool is read-only; use workout_track for every change. A tentative next-set statement alone does not request a workout read; do not call this tool for such statements. Example: {}
workout_get
ChatGPTGet one complete workout by the workout_ref returned from find_workouts or a tracking result: every exercise and set in order, with display loads, notes, and computed insights when available. Use find_workouts to search history; use this only when a returned reference needs full detail. Example: {"workout_ref":"wkt_AbCdEfGh1234"}
workout_record_completed
ChatGPTDraft or confirm an already-completed historical workout. Never use this during a workout currently being performed; use workout_track for live training. Actions: draft, confirm. Example: {"request":{"action":"draft","performed_on":"2026-08-21","exercises":[{"exercise_name":"Bench Press","sets":[{"reps":8,"load_value":60,"load_unit":"kg"}]}]}}
workout_routine_find
ChatGPTList the user's saved workout routines, or read one by routine_ref or by name (matched ignoring case and spacing). A routine is a plan of exercises with optional set targets such as 3 sets of 80 kg x 6-8 and rest, never a record of done work. To train from one, pass its routine_ref to workout_track action=open. Example: {"name":"Push day"}
workout_routine_manage
ChatGPTSave, rename, replace, or archive a saved workout routine, only when the user explicitly asks. Changes apply immediately and return the routine. save_workout turns a completed workout into a routine. Resolve exercises with search_exercises before create or replace_exercises; replace_exercises restates the whole plan. Actions: create, save_workout, rename, replace_exercises, archive. Example: {"request":{"action":"save_workout","workout_ref":"wkt_AbCdEfGh1234","name":"Push day"}}
workout_track
ChatGPTStart or resume a live workout with action=open when the user says 'I just started a workout', 'I'm working out now', or asks to start, resume, or track a workout. Call open immediately, before asking for exercises, sets, or a training location. It recovers an existing active workout instead of creating a duplicate. Do not substitute a workout_current read for this start/resume operation. Do not open a workout for tentative future sets, gym logistics, or travel directions. Also use this tool to reopen the workout the user just finished, prepare completed sets, resolve exercise ambiguity, confirm a prepared change, correct a logged set or exercise assignment, remove a logged set, or finish. Use workout_current, not this tool, to read the current workout without changing it. Use add_sets only for sets the user reports as completed or explicitly asks to log; never use it for planned sets. Note: a bare value reported after a rest usually means a new completed set, not a correction to a previous one. The server chooses the active workout and version. Without workout_ref, changes go to the workout in progress, and corrections fall back to the one just finished. With a workout_ref from find_workouts, add_sets, correct_set, remove_set, and correct_exercise change that finished workout; each such change needs the user's explicit yes before confirm. Follow next_action exactly. Actions: open, reopen, add_sets, correct_set, remove_set, correct_exercise, finish, confirm, select_exercise. Omit exercise_name and exercise_ref to target the current exercise. Example: {"request":{"action":"add_sets","exercise_name":"Bench Press","sets":[{"reps":8,"load_value":60,"load_unit":"kg"}]}}