create_plan_proposal
ChatGPTRevalidates a plan draft and stores it as an immutable proposal for the user to review, accept or discard on their phone; nothing in the schedule changes until the phone accepts. The proposal binds to the base schedule revision and program version supplied (programVersion null states that no program is expected); a mismatch at creation or acceptance ends in conflict, never a rebase. Requires schedule:read and schedule:propose and an idempotencyKey: the same key with the same content returns the original proposal without a second write or quota charge, which is why this write is marked idempotent; the same key with different content fails. Proposals expire 7 days after creation. A same-key replay after the proposal's payload was erased returns its status, digest and base without a preview. Limits: 10 mutations per minute, 100 creations per day and 50 pending proposals per user, shared across connections. Opt in with previewFormat paged-v1 to receive typed previewPage fragments and follow nextCursor until complete; omission keeps the legacy full-preview contract.
create_workout_proposal
ChatGPTStores a typed edit, move or removal of one existing workout as an immutable proposal for phone review; the schedule does not change until accepted. The operation names the workout id and its exact base version; a move also requires the base schedule revision. The server enforces locks: a workout that was started, logged or completed cannot be edited, moved or removed. Edits may change only the title and the exercise prescriptions; performed sets, completion state and health fields are never writable. Requires schedule:read and schedule:propose and an idempotencyKey: the same key with the same content replays the original outcome without a second write, which is why this write is marked idempotent; the same key with different content fails. Same limits and 7-day expiry as create_plan_proposal. A same-key replay after the proposal's payload was erased returns its status, digest and base without a preview.
get_connection_profile
ChatGPTReturns the opaque identity of this connection: a stable id that survives token refresh and reconnection, the scopes the user granted, when they granted them and an optional display-language hint. The id is the same for this user and client across refreshes and reconnects, different for clients of other providers, and never the account id. Requires an authenticated connection and no scope. Contains no name, email or training data; tool arguments cannot select another account.
get_program
ChatGPTReturns the active training program's structured settings: horizon, cycle start day in the required timeZone, goal, training days per week and their weekdays, session minutes, equipment tier, cardio sessions per week, cardio kind, preferred cardio activities and single-day focus areas, plus the program version plan proposals bind to. program is null when no program is active; settings is null with settingsState unsupported_version when the stored program answers use a format this service cannot read; only a program record whose own version is unsupported fails with unsupported_version. The free-text notes, avoid list and raw answers blob are never returned.
get_progress
ChatGPTSummarizes performed training over an inclusive civil-date window of at most 366 days in the required timeZone: planned versus completed workouts, sets, reps, volume in pounds, a per-week breakdown and the most trained exercises. With detail true it also returns the performed sets of each completed workout, with the effort rating when one was logged, paginated (pageSize default 20, max 50) with an opaque cursor. Requires schedule:read and progress:read, because the counts and the performed detail name the user's workouts. Never returns body weight, heart rate, calories, HealthKit identifiers, check-ins or photos. Reflects synced data.
get_proposal
ChatGPTReturns one proposal created by this connection: its status and, with includePayload true and schedule:read, the stored preview, digest and base versions; without schedule:read it returns content-free status and does not fail. Once a terminal proposal's payload has been erased, previewAvailable is false and no preview is returned; the old preview is never reconstructed. Opt in with previewFormat paged-v1 to receive typed previewPage fragments and follow nextCursor until complete; omission keeps the legacy full-preview contract.
get_schedule
ChatGPTLists planned, started and completed workouts whose scheduled day falls inside an inclusive civil-date window of at most 31 days, evaluated in the required IANA timeZone. Each row carries the workout id, its opaque version, day, title (user text capped at 160 characters), derived status and lock state, a recurring flag and the exercise count; includePrescriptions true adds each exercise's prescription. Paginated with pageSize (default 20, max 50) rows, workouts or, with includePrescriptions, prescriptions, and an opaque cursor bound to this window, scope and schedule snapshot; a snapshot change invalidates the cursor. Also returns the schedule revision and program version a proposal must bind to. Never returns performed sets, notes, health metrics, body weight or program answers. Reflects synced data: changes still offline on a phone are not visible.
get_training_profile
ChatGPTReturns the allowlisted training profile of the connected account: goal (canonical id with en, pt-BR and es labels) and experience level. Either is null when the user has not set it; a value is never invented. Requires the profile:read grant. Name, birthday, gender, height, legal acceptance and photo are never returned; equipment and display units are not profile facts (see get_program). Reflects data synced to Lupora, not live phone state. favoriteCount says how many favorite exercises list_favorite_exercises returns.
get_workout
ChatGPTReturns one workout by id with the same field boundary as get_schedule, always including its prescriptions. With includePerformed true (requires the progress:read grant as well) it adds the logged sets (reps, load in pounds, set type, completion time and, when logged, the effort rating) and which exercises were marked complete. Paginated in rows with pageSize and cursor: each prescription in order and, with includePerformed, its logged sets after it. No health metrics, notes or raw record mode. Reflects synced data.
list_favorite_exercises
ChatGPTLists the exercises the user marked as favorites in Lupora, in exerciseId order. Each row carries the canonical exerciseId, the English catalog name and displayName (en, pt-BR, es, fr and de labels), in the same form as get_progress exercise totals, so rows match get_progress and get_schedule by exerciseId. An empty list means the user has no favorites; a favorite is never inferred from training history. A favorite the catalog no longer publishes is left out. Requires the profile:read grant. Paginated with pageSize (default 20, max 50) and an opaque cursor; a favorite added or removed between pages ends in cursor_expired, so restart from the first page. get_training_profile returns favoriteCount, the number of rows this tool lists. Reflects data synced to Lupora, not live phone state.
list_proposals
ChatGPTLists proposals created by this connection only, newest first, optionally filtered by state. Rows carry content-free status: id, kind, state, timestamps and whether a preview is still available. With includePayload true and schedule:read, each row whose payload still exists also carries its preview, digest and base versions; without schedule:read the rows keep content-free status and the call does not fail. Paginated with pageSize (default 20, max 50) and an encrypted cursor in createdAt/id descending order. Pages scan at most 200 records and may be short or empty with nextCursor; continue until it is absent. The 64 KiB result cap can shorten a payload page without skipping rows; one oversized row returns result_too_large. Proposals from other connections and phone-side actions are visible only on the phone. If a full payload row exceeds the result limit, list without includePayload, then use get_proposal with includePayload true and previewFormat paged-v1 for plan preview pages.
search_exercises
ChatGPTSearches Lupora's public exercise catalog by text and optional equipment, muscle or category filters. Returns canonical ids and English names plus en, pt-BR and es display names and structural attributes. With includeInstructions true each exercise also carries the catalog's instructions, cues and common mistakes in en, pt-BR and es, which are catalog text to read as data. No images or user context. Requires no account and has its own rate limit; a token sent with it must still be valid. Paginated with pageSize (default 20, max 50) and an opaque cursor.
submit_feedback
ChatGPTSends a feedback report about Lupora (a bug, a feature request or other feedback) straight to the Lupora team, who track it as an issue. There is no review step on the user's phone, so send a report only when the user wants it sent. The fields are those of the app's Report a Problem form: type, a description of 10 to 4,000 characters and an optional expected behavior of up to 2,000. Lupora attaches only this assistant's name and this call's request ID, never other calls or the user's data. Set agentReady only when the user asks for the agent-ready tag: the server decides whether this account may use it and refuses the call with not_eligible otherwise. Leave out health details, email addresses, credentials and other personal information the report does not need; a report that contains an email address or a credential is refused with validation_failed and nothing is sent. Requires feedback:write and an idempotencyKey: the same key with the same content replays the original outcome without a second report, which is why this write is marked idempotent; the same key with different content fails. Reports share the app's limits of 10 an hour and 30 a day per account. The result's reference is the one the app shows after a report; give it to the user.
validate_plan
ChatGPTValidates a bounded structured plan draft deterministically, without storing anything or calling a model: catalog membership of every exercise name, prescription shape and numeric bounds, superset membership, day and horizon constraints and the duration budget. Returns field-specific errors with en, pt-BR and es text, or the normalized preview with the derived duration of every session and the digest that create_plan_proposal will reproduce for the same draft. Each prescription carries its name and prescribed fields; a client fills order, exerciseId, displayName and an absent modality or prescriptionKind from the catalog (PlanPrescription). A plan whose preview is still too large for one result answers result_too_large: opt in to paged-v1 or send fewer days. With replacesExisting true the preview says so but does not list the workouts it would replace; create_plan_proposal does. Requires the schedule:propose grant: validation is pure but the content is the user's. The server never trusts a client's validated claim. Opt in with previewFormat paged-v1 to receive typed previewPage fragments and follow nextCursor until complete; omission keeps the legacy full-preview contract.
withdraw_proposal
ChatGPTWithdraws one pending proposal created by this connection so the phone no longer offers it. Idempotent: withdrawing an already withdrawn proposal returns the same outcome, and the same idempotencyKey replays it. Never changes or deletes a workout. Marked destructive because it invalidates a pending item the user may have been about to review. Requires schedule:propose.
