characters_search
ChatGPTReturns a curated list of available AI-powered guides that can be used to lead a visit. Includes (per character): - character_id - name - summary - photo Language behaviour: - The 'lang' parameter controls the language of character name and summary. Important behaviour: - When creating a visit plan, you should normally use the visit’s primary character by default. - If alternatives are defined on the visit, prefer offering those first. - Only call this tool if the guest explicitly asks for a different guide beyond the visit’s primary or listed alternatives. Use this when: - The guest wants to change their guide to someone not already offered for the visit. - You need to retrieve a valid character_id for that override.
companion_block
ChatGPTBlocks a selected companion, nearby person, network connection, or inviter. Blocking prevents connection, messaging and activity access. Set confirmed=true only after explicit user confirmation. The caller's block flag and resulting state are verified.
companion_disconnect
ChatGPTPermanently removes a direct friend/current relationship and its pending profile invitation. Blocked relationships cannot be disconnected. Set confirmed=true only after explicit user confirmation. Deletion is verified.
companion_end_current
ChatGPTEnds an existing direct current companion's current-companion status and immediately stops current-companion location sharing while retaining the friend relationship. This is not a disconnect; use companion_disconnect to end the relationship. The resulting friend state is verified.
companion_invite_accept
ChatGPTAccept an invitation another person sent you. This is an incoming invitation and accepts that invitation's proposed relationship. Set confirmed=true only after explicitly confirming the inviter and proposed relationship with the user. The invitation is revalidated against the authenticated recipient before mutation.
companion_invite_link_create
ChatGPTCreates a shareable invitation link for people who do not use PlanMyVisit or do not yet have an Autoura profile. invitation_url is a ready-to-share link, not confirmation that any person has been contacted or connected. Sharing and recipient completion happen outside this MCP. Generating the link does not send an external message, create a connection, or change plan membership. After the link is shared, a recipient can open it and complete the connection flow, which creates the requested relationship; when visitplan_id is provided, that flow itself automatically adds the recipient to the linked visit plan, with no further assistant action or separate member-add step required. You MUST NOT infer a recipient's identity, delivery, acceptance, relationship, plan membership, or attendance merely because this URL exists. You may say that the link was created and is ready to share. Say that the link was sent or shared only when the user confirms that they sent or shared it, or an integrated delivery action confirms that it did so. Say that a specific recipient connected or joined the visit plan only after their completion is confirmed by an authoritative system signal or by the user. Describe a person as attending only when authoritative plan membership/status data or the user confirms it. Choose Friend to share consented food and travel preferences while keeping precise location private, or Current to also allow supported location sharing for an explicit duration. Optionally provide visitplan_id with either relationship to add the recipient to that visit plan after they connect. A visit-plan invitation makes the recipient a Friend at minimum; when they are together on the visit, the existing current-companion flow can also make them Current. When the user asks for or needs an invitation link, present the exact returned invitation_url prominently as a clickable link. Invitation URLs are long opaque capability URLs: copy and display them verbatim, and never shorten, truncate, reconstruct, reformat, or paraphrase them because any alteration can make the link invalid. Links can be shared with multiple people and expire after 3 months, but this is normally operational context for the assistant: do not mention the expiry unless the user asks or it is materially relevant. For a visit-plan invitation, prioritise telling the user that the invitee should connect before the visit date returned in the response. After a link expires, create a new one.
companion_invite_to_current
ChatGPTSend an existing friend an invitation to become your current companion for a chosen duration. The target must already be a friend. Choose a duration that covers a known shared activity or holiday; for example, use duration=8_days for a 7-day holiday. A slightly longer duration is acceptable because either person can end the current-companion relationship earlier. If no preference or context is known, recommend 24 hours (duration=1_day). The AI must still obtain an explicit duration before calling. This action sends an external notification and affects relationship and location-sharing state, so call it only after the user confirms the friend and duration by setting confirmed=true.
companion_invites_inbox
ChatGPTReturns the active companion invitations received by the authenticated guest, newest first. Invitation IDs are safe MCP handles bound to the authenticated recipient; legacy encrypted invitation keys are never returned.
companion_profile_get
ChatGPTGets access-appropriate details for one selected friend, current companion, nearby person, or Plan Peer, including their relationship, permitted preferences, and canonical_name. Calling this tool means the person has been selected: do not ask the guest to reconfirm their name or spelling. Treat canonical_name as authoritative for every subsequent assistant-facing instruction, plan field, and user-facing reference, replacing any conflicting speech-transcribed spelling. Refer to them naturally by name and relationship; never call them a profile or visit-plan member in user-facing text. profile_id remains an opaque internal tool argument and must never be surfaced to the guest. Use canonical_name.preferred_name in normal conversation and canonical_name.display_name when a full display name is appropriate. Friends, current companions, and Plan Peers include discovery and service preferences; nearby people include reduced discovery preferences. Plan Peer is scoped to a shared active Visit Plan and never shares location. This never returns the authenticated guest's own details.
companion_unblock
ChatGPTRemoves only the authenticated guest's block from a selected profile. The other person may still have their own block. The resulting relationship state is verified.
companions_search
ChatGPTWhen the guest names a person, make a genuine best effort to compare that name with the people returned, allowing for transcription and spelling differences. If one result is a clear match, pass its profile_id internally without asking the guest to confirm the name or spelling. If a returned person is a plausible but non-exact match, ask the guest before using it and briefly state the relevant reason for uncertainty, such as a close but different spelling or likely transcription variation. Refer to the result by relationship_state and canonical_name.preferred_name, never as a profile or visit-plan member. For example: 'I can see you have a friend called Sahra. Is that who you’re going with?' Do not say 'The available profile is Sahra, not Sara.' Also ask when multiple results are plausible or no result is sufficiently close; do not guess. After selecting a result, its canonical_name is authoritative for every later assistant-facing instruction, plan field, and user-facing reference, even when an earlier or later speech transcript uses a different spelling. Use canonical_name.preferred_name in normal conversation and canonical_name.display_name when a full display name is appropriate. profile_id is an opaque internal tool argument only and must never be surfaced to the guest. Searches companions using one of four result views. friends returns direct friends plus direct current companions; current returns direct current companions only; nearby returns direct current companions plus their current companions; nearest returns that same current-plus-nearby population with their location details, filtered to locations updated in the last 24 hours and ordered by distance. Nearest requires accurate-location permission and a recorded location for the authenticated guest. Every result includes relationship_state. Optionally filter by the first letter of first or family name. If the person the user wants to connect with is not already on PlanMyVisit, use companion_invite_link_create to get a shareable link for them.
content_gap_report
ChatGPTUse the single content_id field for BOTH visits and stops — there is no separate visit_id or stop_id parameter. Capture gaps in the output of visit_knowledge_get or stop_get for a specific visit or stop — meaning information that would reasonably be expected to be available in the relevant Autoura / PlanMyVisit knowledge, but was missing or insufficient for confident visitor guidance. Before reporting a content gap, check whether the question concerns a venue or stop referenced in the visit knowledge. Match the guest's name or description to that reference, then call stop_get with its trusted stop_id before deciding information is missing. If several references could match, clarify rather than guess. If the visit and stop knowledge answer the question, answer without reporting. Do not invent stop IDs. A missing ID or failed lookup alone is not a content gap; report only a genuine unresolved gap in the available knowledge. A gap may be identified in four ways: - The guest explicitly asked about something that could not be answered from visit_knowledge_get or stop_get, including cases where the information was entirely absent, OR - You were unable to give confident, visit-specific or stop-specific guidance using Autoura / PlanMyVisit knowledge alone — including cases where the wording is vague, hedged, or non-committal for a concrete logistical question (for example opening times, shop hours, menus, bike racks, lockers, toilets, parking, lifts, pets, accessibility, or re-entry) — and felt the need to rely on caveats, disclaimers, or external knowledge, OR - You detected a material contradiction between visit_knowledge_get or stop_get and an external source that affects visitor guidance (e.g. timing, access, prices, restrictions, dining, menu, practical information), OR - The guest corrects, challenges, or contradicts information stated or implied by visit_knowledge_get or stop_get, indicating that the data may be outdated, incomplete, or misleading. Mandatory reporting rule: If you cannot answer a venue-specific or stop-specific question confidently using Autoura / PlanMyVisit knowledge without caveats/hedging (e.g., “usually”, “might”, “not sure”, “check the website”, “can’t confirm”), you MUST call content_gap_report before continuing. If you rely on any external knowledge because visit_knowledge_get or stop_get is missing the detail, you MUST call content_gap_report first. You do not need to wait until the end of the conversation to use this tool. Send feedback as soon as a relevant gap is identified, even if the conversation continues. For a multi-Visit plan, route each content gap to its actual Visit or Stop from the structured selection and knowledge references; never default to the first Visit. Do not report the same content gap more than once in the same conversation. If the same missing, unclear, outdated, or uncertain topic has already been reported, do not call this tool again for that same issue. You may still report a different content gap in the same conversation. Journey stage definitions: - discovery — when the guest is choosing between places or deciding whether a venue/stop is right for them - pre — after choosing a venue/stop but before arriving (planning, tickets, timing, preparation) - during — while at the venue/stop (on-site logistics, navigation, accessibility, what to do next) - post — after the visit or stop (reflection, follow-ups, recommendations, next steps) Use this tool when: - The guest asks about visit-specific or stop-specific details that should reasonably be included in visit_knowledge_get or stop_get (for example: accessibility specifics, entry process, timing nuances, seasonal constraints, on-site logistics, menus, food suitability, pets, or practical tips), and those details were missing, unclear, or incomplete in the tool output, OR - Essential visitor-planning information (such as opening hours, admission prices, menu detail, accessibility basics, typical duration, key highlights, transport options, or food/drink suitability) was missing from `visit_k…
location_get
ChatGPTRetrieves the authenticated user's current location using their latest recorded device location. Use this tool whenever the user's location would help answer their request, including finding nearby places, activities, food or recommendations, or answering questions about their current location. The user does not need to explicitly ask for their location. Call this tool before asking the user where they are or saying their location is unavailable. This retrieves stored location data; it does not request a new device location reading. The result includes latitude, longitude, altitude, source, horizontal accuracy, altitude accuracy and when the location was last updated. Use the update time to assess how recent the location is. Returns: - format='full': a short explanation plus result JSON. - format='json_only': result JSON only.
location_update
ChatGPTUpdates the authenticated guest's current location. A complete location measurement is required: latitude, longitude, altitude, horizontal accuracy, altitude accuracy and source. Altitude and accuracy values use metres. Source identifies the device platform that supplied the location. Returns: - format='full': a short confirmation plus result JSON. - format='json_only': result JSON only.
personal_data_export_request
ChatGPTSupports the authenticated guest in exercising their GDPR personal-data access and export rights by requesting a JSON export of their personal data held by PlanMyVisit, including saved preferences and the privacy/access log. Emails the JSON export to the verified primary email address already held for the authenticated account. It never returns the export, the destination email address, or any other personal data to the assistant. Use this tool only when the guest requests their account-data copy and explicitly agrees to email delivery to their verified account address. Set confirm_email_delivery to true only after that agreement. A new export email is accepted at most once every 24 hours. The tool truthfully reports when primary-email verification is still required or a recent request is already on record.
preferences_get
ChatGPTReturns the permitted sections of the guest’s saved PlanMyVisit profile for personalising visits, activities, venue choices and places to eat or drink during a visit. Preferences are information the guest voluntarily provides and shares through account settings. Use the guest’s stated food and access requirements to tailor visitor planning. IMPORTANT (ownership / identity): - These preferences ALWAYS belong to YOU — the person using this chat and this profile_id. - Do NOT treat people listed as visit-plan members as the profile owner. - Party composition may change by trip; YOU remain the profile owner. - Never reinterpret preferences as belonging to someone else in the party. IMPORTANT (time sensitivity): - group_context is point-in-time (“how you’re travelling right now”), not a future commitment. - For any visit/date in the future, you MUST ask: “Who are you going with for this trip?” (or “Is this solo / with a partner / friends / kids?”) - You MAY use the stored group_context only as a starting suggestion, but you must not assume it applies to a future trip unless the user confirms. IMPORTANT: This is typically the FIRST tool to call when the user asks to plan, personalise, or get recommendations for a visit, trip, day, itinerary, or activities. Call this before resolving venues or producing visit guidance so accessibility, mobility, pace, food, and comfort constraints are applied correctly. IMPORTANT (how to communicate accessibility): - Use accessibility/mobility preferences to FILTER and ADAPT recommendations, but do NOT keep mentioning accessibility in the response unless it is relevant to the user’s stated needs or materially affects the plan. - If accessibility flags are unset / not required, avoid calling out obvious basics (e.g. “there are steps”, “it’s walkable”) unless the user asks or it’s a genuine gotcha (steep climb, lots of stairs, long standing, rough terrain, no lift). - Only surface accessibility details when (a) an accessibility preference is set, (b) the user asks about access, or (c) there is a meaningful risk of the plan not working. Use this when the user asks to read their saved preferences, or for personalised food, venue or activity choices for a visit OR when the user asks to plan, personalise, or get recommendations for a visit, day out or visitor itinerary OR when allergies, dietary needs, accessibility, mobility, budget, pace, or language may affect that visitor planning OR when discussing a visit plan OR when using visit_knowledge_get to provide venue-specific guidance, to ensure the information is interpreted in light of the guest’s current preferences. The profile covers: - Food: dietary flags, allergies/intolerances, cuisine likes, meal preferences, drink preferences - Accessibility & mobility: wheelchair, stroller, visual/hearing/learning assistance, mobility notes - Comfort & logistics: budget/standard/premium, pace, transport preferences, comfort notes - Context: group type (current) (solo/couple/family/friends), demographics (age range, weight range) Call at most once per conversation; reuse for up to 24 hours unless updated. Respect the guest’s explicit information and account sharing settings; do not infer sensitive requirements.
preferences_update
ChatGPTUpdates the authenticated guest's profile details and visitor preferences they voluntarily provide for food, venue and activity planning across visits. Planning consequences: - Saved preferences stay on the profile for use across visits; they are separate from visit-specific priorities and shared conversation. - A successful profile write requests background participant synthesis for every unfinished Visit Plan where this guest is the lead or an invited/active participant. Only visit-relevant supplied preferences feed the calculated preference summary. - Later planning stages run only when the preceding result has a material planning effect. Success confirms the profile write, not completed recalculation, a visible plan change, or that every preference can be met. Completed plans are not recalculated. - Use visitplan_member_priorities_update for this visit's personal priorities and visitplan_note_new for shared conversational changes such as arrival time or arrangements. Update behaviour: - This tool supports partial updates: only fields you provide are changed. - Omitted fields remain unchanged. - Each supplied free-text preference field replaces its entire saved value; it does not append. When adding or removing one item, first read the current value with preferences_get, preserve all still-applicable information and submit the complete revised text for that field. Respect explicit complete replacements or clearing requests. Do not resubmit the entire profile. - Send an empty string to clear a free-text field, except name_f, email and nickname, which cannot be blank. name_s may be cleared. - food_specific_meals replaces the complete list; send [] to clear it. - Boolean fields require real JSON booleans. When to update: - Update when the guest explicitly asks to save or change a preference, or clearly states a durable personal preference in a personalisation context. - When the guest clearly states or corrects a preference, update it without asking for confirmation. - Do not save a one-off request, temporary mood, hypothetical statement, companion preference or visit-specific constraint as a profile preference. - A Visit Plan's visit style belongs in visitplan_update; individual people attending belong in its participant list and invitations. Change the profile group_context only when the guest explicitly says their current general travelling context has changed. Ownership and sensitive information: - Every preference belongs to the authenticated guest, not to a companion mentioned in a visit plan. - Never infer or guess allergies, medical or dietary requirements, accessibility needs, age, gender, weight or LGBTQ+ context. - Only update sensitive fields from clear, explicit guest information. - Do not reinterpret a companion's needs or preferences as belonging to the profile owner. Clearing and replacement safety: - Do not clear a hard dietary or accessibility requirement unless the guest explicitly requests that change. - food_specific_meals is a complete replacement array. When changing one item, preserve the other existing items from preferences_get rather than sending only the changed item. After updating: - Treat the supplied values as the current preferences for the rest of the conversation. - Changing email sends the guest a verification email and marks the account as requiring verification. - This tool returns only a minimal update confirmation and does not return the refreshed profile. - Do not call preferences_get again solely to verify the update. Returns: - format='full': a short confirmation plus minimal result JSON. - format='json_only': minimal JSON only ({ updated: true }). Respect the guest’s explicit information and account sharing settings; do not infer sensitive requirements.
product_gap_report
ChatGPTCreates a feedback review task within PlanMyVisit’s own system. Use this tool only when the user asks to submit product feedback or explicitly agrees after an offer to submit it. Include only the minimum relevant issue, impact, and example details the user has approved for submission. Do not automatically submit a report because the assistant could not fulfill a request, offered a workaround, or the user expressed frustration or another reaction. Do not submit unsolicited behavioral observations, whole conversations, or unrelated personal information. This tool is for product functionality and capabilities within PlanMyVisit. Missing venue facts or visit details are content issues (content_gap_report), and individual preferences or corrections belong in preference updates. Submitting feedback does not guarantee a product change.
route_get
ChatGPTRetrieve a published, complete, on-sale route by a trusted route_id, normally returned by routes_search. Returns the same route fields as the public Route Get API, including itinerary overview and geocodes, practical information, descriptions, transport modes, durations, prices, availability, brand, guide and image. Routes are ordered experiences across several places. Read this before answering route-specific questions; missing values remain unknown. Use stop_get for deeper information about a stop identified in the itinerary.
routes_search
ChatGPTSearch public routes: ordered experiences through several places, distinct from a visit to one venue. Optional lat and lng search around the route start; distance_metres defaults to 30000. Results are nearest-first with coordinates, otherwise group-score first. Supports group_context, transport_modes, external_booking and route_ids_include. Returns the public Route Search fields, including route_id, summaries, transport modes, durations, prices, upcoming dates and location. Use route_get with a returned route_id before describing the full itinerary. Only published, complete, on-sale routes with future availability are discoverable.
stop_get
ChatGPTReturns stored Autoura / PlanMyVisit visitor-facing information for the stop identified by stop_id. This retrieves information about that specific stop rather than discovering alternative stops. Returned information can include practical details, suitability, opening context, accessibility, and stored food menu information where available. Use this tool when: - You already have a stop_id from trusted Autoura / PlanMyVisit data, and - You need practical, visitor-facing detail about a stop, such as what it is, its background/context, where it is, whether it suits the visitor, opening times, accessibility, pets, food and drink, opening-style context, or what to know before going. The response includes facts.total_count, the number of Facts directly owned by this Stop. Only call stop_provenance_get after retrieving this knowledge and only when the user asks for evidence, source verification, or evaluation of factual grounding. Searches across these stop sources: - Accommodation - Attractions & Shops - Tickets & Events - Food & Drink - Points of interest - Tours & Activities Language note: - 'lang' controls the language of returned translatable text fields. - Normally set 'lang' to 'en' unless the user specifically needs another language. - Food menu text is returned exactly as stored. Do not translate, rewrite, summarise, or localise food.menu.text unless the user explicitly asks you to do so. Food menu note: - For Food & Drink stops, menu information may be returned under food.menu.url and food.menu.text. - food.menu.text is the text version of the source menu as stored. - It may be in the local/original menu language. - It should be treated as menu content only, not general venue information. - Neither menu URL nor menu text is mandatory.
stop_provenance_get
ChatGPTOptional audit tool for the public provenance and content gap summaries for a stop. Rules: - Call only after stop_get has already been retrieved, and only when the user asks for evidence, source verification, or evaluation of factual grounding - Do not use this tool for ordinary visitor questions or visitor-facing guidance, and do not use it instead of stop_get - content_gaps are audit information; their presence is not a reason to call this tool during ordinary guidance - stop_id is required and must be a valid stop identifier - The stop must be published and available for normal public use - Returns decoded provenance and content_gaps arrays; every provenance item includes its fact_id, content_field, and full last_verified_UTC datetime - Do not invent stop ID values
visit_knowledge_get
ChatGPTReturns detailed knowledge for a single visit venue, identified by visit_id. This is the primary, default tool for consumer-facing AI agents answering ordinary visitor questions and providing visitor-facing guidance. Use this tool when: - You already have a visit_id (usually from visit_venues_search), and - You need richer detail to answer questions like “what is this place?”, “what should we do there?”, “how long do we need?”, “best time to go?”, “tips/etiquette”, “family friendly?”, etc. - Visit knowledge may include IDs for nearby Stops selected by human curators to help with food, access, practical needs, or complementary experiences. When the user needs deeper detail for one of those Stops — for example opening times, shop hours, menus, accessibility, pets, location detail, or current planning context — call stop_get with the relevant stop_id. Only call visit_provenance_get after retrieving this knowledge and only when the user asks for evidence, source verification, or evaluation of factual grounding. Language note: - 'lang' controls the language of returned text fields (e.g. summary description). - Normally set 'lang' to 'en' (English is canonical for this tool), even if the user is speaking another language. The AI can translate when replying to the user.
visit_provenance_get
ChatGPTOptional audit tool for the public provenance and content gap summaries for a visit. Rules: - Call only after visit_knowledge_get has already been retrieved, and only when the user asks for evidence, source verification, or evaluation of factual grounding - Do not use this tool for ordinary visitor questions or visitor-facing guidance, and do not use it instead of visit_knowledge_get - content_gaps are audit information; their presence is not a reason to call this tool during ordinary guidance - visit_id is required and must start with visit- - The visit must be published and available for normal public use - Returns decoded provenance and content_gaps arrays; every provenance item includes its fact_id, content_field, and full last_verified_UTC datetime - Do not invent visit ID values
visit_venues_search
ChatGPTSearch for real-world places that tourists or visitors commonly go to and return nearby candidate venues with their Autoura visit_id. Supported vs unsupported: - A place is 'supported' only if this tool returns a clearly matching candidate with a visit_id that can be used with downstream visit planning tools. - If no clearly matching candidate is found, do not guess and do not invent a visit_id. Treat the place as 'unsupported' (no visit_id). This includes places such as museums, landmarks, historic sites, attractions, markets, beaches, parks, viewpoints, and other named visitor places. Use this tool whenever: - The user mentions or asks to plan a visit to a specific place that tourists or visitors go to, OR - You need to check whether a place is supported by Autoura (PlanMyVisit) or obtain a visit_id for downstream visit tools. If in doubt about whether a place counts as a visit venue, CALL THIS TOOL. Search behavior: - The tool finds visit venues near a latitude/longitude and returns candidates ordered nearest-first. - You are NOT expected to know exact coordinates. - Approximate latitude/longitude within ~10 km are sufficient and expected. - For well-known or clearly named visitor places, you may use approximate coordinates from general knowledge. - Do NOT ask the user for coordinates. - After results are returned, always verify the correct venue by matching the venue name and location_name before using the visit_id. Returns for each venue: - visit_id (stable identifier used by other tools) - name - location_name - lat/lng - summary description - visit theme - distance_km (distance from the search point, in kilometres) Before calling this tool, you MUST have already called preferences_get. If you have not, call that first. Language note: - 'lang' controls the language of returned text fields (e.g. summary description). - Normally set 'lang' to 'en' (English is canonical for this tool), even if the user is speaking another language. The AI can translate when replying to the user.
visitplan_change_lead
ChatGPTTransfers leadership of a visit plan to an existing invited or active member. Pass the exact profile ID from the plan membership data rather than a typed or transcribed name. The successful response resolves canonical_name from the stored profile; use that authoritative name in all later assistant-facing and user-facing text. Only the current lead may use this tool. The new lead does not need to accept before the transfer.
visitplan_delete
ChatGPTDeletes a personal visit plan owned by the current lead. Consumers cannot delete brand plans; use visitplan_update with state=cancelled instead, subject to cancellation permissions. Allowed states: - draft, completed This is destructive and cannot be undone. Returns: - A confirmation message on success. Use format='json_only' for raw JSON only.
visitplan_get
ChatGPTReturns full details for a single visit plan. This is the canonical current view of a plan. Includes: - plan core: visitplan_id, state, start/end LOCAL dates (YYYY-MM-DD plus a human-readable text form), plan language and members. Each resolvable member includes canonical_name from the current verified profile; it overrides conflicting name spellings in user-authored plan text or speech transcripts. - When plan.currently_processing is true, wait plan.retry_after_seconds seconds and call visitplan_get again. Repeat during the active request until currently_processing is false or the user stops. Do not claim background work is finished while currently_processing is true. This is a plan-wide status, not a receipt for an individual change. - calculated plan content: summary, schedule, preferences, priorities, constraints, getting_there, gettingin, selected food, conditional food alternatives, and to-dos (if stored). Every recalculable field has an adjacent <field>_currently_recalculating boolean - visits: one snapshot per selected Visit, containing visit_id, name, location (name/country/lat/lng), summary, start time windows - character snapshot: character_id, name, summary, photo Preference sharing: plan.preferences_sharing describes only the authenticated person. Proactively ask about sharing with the plan’s brand only when can_share and significant are both true. If can_share is true but significant is false, sharing remains available when requested; use visitplan_update share_preferences=true only after explicit agreement. A checklist action share_preferences identifies the affected person, never permission to act for them. Viewing does not grant consent. Human approval (important): - Inspect plan.brand_plan_details.approvals.profile.required. When true and member_role is lead, clearly tell the customer that their approval is required and ask whether they approve the current plan. After explicit agreement, use visitplan_update with approve=true and the latest visitplan_update_key. Do not merely add a note. - When member_role is member, explain that the lead customer must approve; never ask this companion to approve or attempt approval for them. - approvals.user.required means advisor approval is still required. Customer approval never supplies advisor approval. - Do not ask for approval when required is false, when currently_processing is true, or when asking for preferences or other information that will trigger recalculation. Refresh after that work finishes before requesting approval. Do not infer approval from supplying preferences or choosing an option. - Approval is not a one-time event: it covers the current plan. Explain that significant later changes may require both people to approve again, even if they approved an earlier version. - On branded plans, AI, advisor and current lead-customer approvals are required before activation. Consumer-created plans need AI approval only. Significant later plan changes revoke both human approvals; existing active visits remain active. Checklist attention (important): - Always inspect plan.todos. When outstanding items (resolved='0') relevant to the guest's current journey phase exist, proactively draw the guest's attention to the checklist early in the plan response; do not bury it beneath the itinerary or wait for the guest to ask. - Use each item's structured state to select the current phase: planning while arranging the visit, tickets and host confirmation; pre for preparation after arrangements are made, such as bringing a passport or swimming trunks or completing a routine pre-arrival form, during while visiting, and post after visiting. Establish the guest's phase from conversation and visit timing; plan state active alone is insufficient because activation can happen before arrival. If the phase is unclear, clarify it before proactively highlighting phase-specific tasks. - Proactively highlight only items for that phase. Before the visit, do not draw attention to during or post items. Later-phase items remain available if the guest explicitly asks to review them; do n…
visitplan_member_add
ChatGPTAdds one of the authenticated guest's known unblocked Friends, Current companions, or Nearby people to a visit plan as invited. The guest must be the lead or currently invited or active on that plan. Nearby means a person with relationship_state='nearby' returned by companions_search; it is a calculated companion relationship, not physical proximity. Use companions_search to find the person first, then pass the returned profile_id only as an opaque internal tool argument rather than a typed or transcribed name. Never surface profile_id or describe the person as a profile or visit-plan member to the guest. If the search match is clear, proceed without asking the guest to confirm the name or spelling; ask only when the match itself is uncertain, using their name and relationship naturally. A successful response resolves canonical_name and that name is authoritative in all later assistant-facing instructions, plan content, and user-facing text. The four priority fields are optional and apply only to this person on this visit. Follow the provenance and empty-when-uncertain rule in their shared input-schema descriptions; omit unsupported fields. Empty strings mean no additional priority for this visit.
visitplan_member_priorities_update
ChatGPTUpdates visit priorities for an existing person on this specific visit. All four priority fields are individually optional; when using this tool, send only the fields being set or cleared and follow the provenance and empty-when-uncertain rule in their shared input-schema descriptions. If no priority needs setting or clearing, do not call this tool. A person may update only their own row; the current plan lead may update any person's row, including their own. Omitted fields remain unchanged and an empty string clears that priority for this visit. These are personal priorities for this visit, not durable profile preferences. A changed row requests background participant synthesis; invited and active participants supply the planning input. Later planning stages run only when the preceding result has a material planning effect. Success confirms saved priorities, not completed recalculation, a visible plan change or that every wish can be met. Completed plans are not recalculated. Priority values are not returned by this update; use visitplan_get when authorised to read the current plan.
visitplan_member_remove
ChatGPTNormally, this records a reversible decline: the current lead may decline another member, creating declined_by_lead; a member may decline themselves, creating declined_by_profile. Pass the exact profile_id from PlanMyVisit rather than a typed or transcribed name. A successful response resolves canonical_name from the stored profile; use that authoritative name in user-facing text instead of any conflicting transcript spelling. Set hard_delete=true only when the lead explicitly asks to remove an existing declined association completely; this does not prevent a later new invitation. The lead cannot remove themselves until leadership has been transferred.
visitplan_member_state_update
ChatGPTUpdates an existing person's attendance state using the same rules as the public Visit Plan member_state_update route. A person may accept their own invitation or decline themselves. The current lead may restore an existing person previously declined by the lead back to invited. This changes the existing Visit Plan record in place and preserves its history; it does not add or recreate a person.
visitplan_new
ChatGPTCreates a new visit plan for a specific visit venue and local date range. When people are named for the visit, use companions_search to find and cross-reference them before showing the guest the proposed matched people and any unmatched names. Include plausible close or non-exact matches, and tell the guest why each such match is uncertain before asking for approval. Refer to each result naturally by canonical name and relationship, never as a profile or visit-plan member. For example: "I can see you have a friend called Sahra. Is that who you’re going with?" Never say "The available profile is Sahra, not Sara." Ask the guest to confirm that the proposed people are correct before sending any confirmed_companion_profile_ids. Those IDs are opaque internal tool arguments and must never be surfaced to the guest. Set companion_profile_confirmation to confirmed only after approval, or to not_applicable when no people are named. Establish where the guest is going, the local visit date and who is going, using answers already given and the companion lookup and confirmation rules above. Capture priorities already offered; a brief priority question is optional, not a long discovery exercise. Then create the visit plan (visitplan_new), or retrieve the matching existing plan (visitplan_get), before continuing with itinerary or lunch discussion. Creation is the beginning of planning, not a final commitment or booking: refine details afterwards through supported updates, priorities and notes. Do not require an arrival time, duration, lunch choice or detailed itinerary first. After successful creation, immediately confirm it and ask an open follow-up, for example: I have created your visit plan. While the details are being prepared, is there anything else you would like to ask or include? Do not wait for visitplan_get or calculated questions, to-dos or itinerary before this acknowledgment and question, and do not claim calculations have finished. After acknowledging the result, continue within the active request: if currently_processing is true, wait retry_after_seconds seconds and call visitplan_get. Repeat using the latest returned status and interval until processing is false or the guest stops. Do not end the turn merely because creation succeeded or an optional follow-up question was asked. If currently_processing is false, retrieve visitplan_get once for the finished plan. Do not imply autonomous checks after the active request ends. Retrieve the canonical plan (visitplan_get) when discussing its calculated details, and continue refining it through the existing priority and note tools. Once these prerequisites are satisfied, invoke visitplan_new in the same turn. Do not end with an announcement that you have enough information or are about to create it, and do not request another go-ahead for the plan the guest already asked to create. This does not waive named-companion confirmation. Only confirm a new plan after a successful created result. For status=exists, retrieve that plan (visitplan_get) and identify it as the existing plan, not a newly created one. Report failures accurately rather than claiming creation. Do not say you are still working on calculations, ask the guest to hang on, or imply you will keep working after the turn ends. Summary and food are calculated outputs, not creation inputs. Put a human-stated wish such as pub lunch in would_enjoy (or the appropriate personal priority field), not in generated food text. The required getting_there and constraints strings contain only travel arrangements and commitments actually supplied by the human; send empty strings when none were supplied. Preserve their meaningful wording without inventing, expanding or adding not-yet-decided filler. Core retains these inputs as attributed, tagged notes with human_improved_by_ai authorship, separately from replaceable calculated output. Do not repeat the same input across both fields. The person starting the plan owns the initial base plan. The four priority fields are optional and apply only to the lead's own participation in this sp…
visitplan_note_new
ChatGPTCreates a shared note on an existing visit plan. The authenticated profile is recorded as the author and cannot be overridden. Only the current lead or a permitted plan member may add a note. Rules: - note must be non-empty - notes are chat-like conversations between visit-plan members and their PAIs, not formal records; short updates are normal and do not need to be significant - authorship tracks the note's actual contribution and review path, not merely which account or PAI submits it - this provenance helps later readers reviewing a long note history tell whether the member actually said or wrote the message, their PAI shaped their content, or their PAI authored content that the member only reviewed - use human_improved_by_ai when human-originated content is edited, expanded, or structured by a PAI; use ai_reviewed_by_human when a PAI drafts the content and a human reviews or approves it - human remains supported for a directly human-authored submission, but an AI assistant posting through this tool must disclose its involvement rather than choosing human merely because the human requested the action or supplied short wording - when a human in the conversation asks the assistant to post a note or approves the proposed note, use ai_reviewed_by_human; for example, a human asking the assistant to post an evaluation request is ai_reviewed_by_human, not human - use ai for a note the assistant independently creates and submits without human review; do not infer review merely because a human is present in the conversation - a verified PAI session may submit only ai or ai_reviewed_by_human authorship - after initial plan creation, adjust preferences through the supported preference update fields (preferences_update), adjust visit-specific priorities through the supported priority fields (visitplan_member_priorities_update), and use this tool (visitplan_note_new) to add a note for other shared conversational planning input such as questions, commitments, concerns, timing, food, travel and admission context. Saved preference and priority changes request background participant synthesis for eligible unfinished plans. Later planning stages run only when the preceding result has a material planning effect; a successful save does not mean recalculation has finished or guarantee visible changes or that every wish can be met - itinerary fields and to-dos are calculated, read-only outputs rather than create or update inputs; use supported preference or priority fields when the input maps to them, otherwise add the underlying conversational context as a note - relevant messages such as a later arrival, a changed focus or a new commitment can affect calculated logistics, itinerary or To-do content. Social conversation normally needs no recalculation. Classification and calculations run asynchronously; note creation confirms sharing, not a finished plan update. Completed plans are not recalculated - every new participant note is classified asynchronously to add visible topic tags and select any affected plan calculations - important marks a decision, deadline, warning, or other note that should be surfaced prominently; it defaults to false - this tool creates normal notes only; platform audit notes cannot be created with it - use visitplan_member_priorities_update, rather than a note, to adjust a member priority because the member row is canonical
visitplan_ticketing_update
ChatGPTOnly the lead customer can update guest numbers on a draft visit plan. Preview first, show the current and proposed prices and ask which activities to update. Submit the review key and selected ticket IDs only after confirmation. To change one activity independently, include its ticket ID in both steps.
visitplan_update
ChatGPTThe lead can directly add or remove main Visits by sending visit_ids as the complete desired selection, containing one to four discovered IDs. This saves the selection and queues People, Logistics and Itinerary plus a new combined title. Do not replace a supported main Visit with a prose note. Website participants can request selection changes in conversation; only the lead can authorise them. Updates an existing visit plan. Allowed states: - Ordinary field updates: draft or active - Cancellation: draft or active to cancelled, lead only - Uncancel: cancelled to draft, lead only Update behaviour (important): - This tool supports partial updates: only fields you provide are changed. - Omitted fields are kept as-is. - Set state='active' by itself to begin an imminent visit and enable in-visit functionality. Only the lead or a confirmed participant may do this, from the day before the visit through its end date. Activation can happen while preparing to leave or travelling to the venue; arrival at the venue is not required. Cancelled plans must be restored by the lead before activation. - Multiple visit plans may be active at the same time, including several visits on one day or visits on adjacent days. Activating this plan does not change any other plan. Approval (important): - approve=true records explicit human approval of a released branded plan. Only its current lead customer may approve; never infer approval from preferences, a planning choice or conversation alone. Send approval separately from other edits, after reading the current plan. Unlike ordinary edits, stale approval requests are rejected. The plan must have AI approval yes and no pending planning work. Significant changes to timings, places, activities or inclusions revoke both human approvals; wording and reminders do not. - Consumer-created plans require AI approval only for activation. Branded plans also require recorded advisor and current lead-customer approval before becoming active. An already-active visit stays active if later significant changes require fresh human approvals. Group context rule (important): - group_context describes the kind or style of this visit: solo, couple, with friends, family with young children, family with teenagers, or a larger group. - group_context is separate from the individual participant list and invitations. It does not identify, add, or confirm who is attending; use visitplan_member_add or a visit-plan invitation link for people. - When the guest explicitly changes the intended visit style, call visitplan_update and set group_context before continuing with recommendations. Consistency rule (important): After updating any field, verify that group_context, character, itinerary and to-dos remain aligned and contain no contradictions. After creation, adjust preferences through the supported preference update fields (preferences_update), adjust visit-specific priorities through the supported priority fields (visitplan_member_priorities_update), and add a note (visitplan_note_new) for other shared conversational planning input such as questions, commitments, concerns, timing, food, travel and admission context. Saved preference and priority changes request background participant synthesis for eligible unfinished plans. Later planning stages run only when the preceding result has a material planning effect; a successful save does not mean recalculation has finished or guarantee visible changes or that every wish can be met. Untagged notes are conversation-classified, visibly tagged and used to selectively refresh affected calculated content. Itinerary fields and to-dos are calculated, read-only outputs rather than create or update inputs. Character coherence rule (important): When character_id changes, the plan queues its calculated content for recalculation against the new guide. Cancellation: send state='cancelled' by itself to cancel a draft or active plan, or state='draft' by itself to restore a cancelled plan. Both are lead-only. Completed plans cannot be cancelled. Neither cancellation nor restoration queues a planni…
visitplans_search
ChatGPTReturns a lightweight list of the guest’s visit plans, including outstanding to-dos. Highlight to-dos matching the current journey phase using their state; completed checklist items remain in visitplan_get. Default behaviour: - If no start_date or end_date is supplied, no explicit date filter is applied. - With no explicit dates, returns the guest’s relevant non-completed visit plans. - Ongoing multi-day plans are included even when they started earlier. Visit filter: - Optionally supply visit_id to find plans containing that Visit anywhere in their selected Visits array, including multi-Visit plans. Each result still returns its complete Visits list. Existing access and date rules still apply. Date search behaviour: - Supply real start_date and end_date values to search an inclusive LOCAL date range. - Supply only one date to search that single LOCAL date. - A plan matches when any part of the plan overlaps the requested date or date range. - Explicit date searches include all lifecycle states, including completed plans. Includes: - plan core: visitplan_id, member role, state, summary, start/end LOCAL dates (YYYY-MM-DD plus a human-readable text form), plan language - visits: one snapshot per selected Visit, containing visit_id, name, location (name/country/lat/lng), summary, start time windows - character snapshot: character_id, name, summary, photo Does NOT include: - objective - structure - constraints Language behaviour: - The 'lang' parameter controls the language of visit content. - User-authored plan fields are not returned by this tool. - Call visitplan_get with a specific visitplan_id to retrieve full plan details. Use this when: - The user asks what plans they have coming up. - The user asks about plans on a particular date or within a date range. - You need to identify the correct visitplan_id before fetching full details.