display_recommendations
ChatGPTSend a curated selection to the Tawadoo widget. Search tools are data-only and do not render the widget. Call this tool exactly once after a final relevant search result, using the exact product_url values of the listings you will recommend in text. The widget will display those listings as product cards — nothing more, nothing less. When to call - A search returned relevant recommendations and you have picked your 1–10 final recommendations → call this exactly once before answering - Multiple relevant searches were needed and you have picked your 1–10 final recommendations → call this exactly once before answering - One broad search returned many relevant listings and you want the widget to show only your final subset → call this exactly once before answering - is_relevant=false or no recommendations → do not call this (nothing to display) What to pass - product_urls: the ordered list of product_url values of the listings you are recommending, best match first - Pass exactly the URLs you will link in your text response — the widget mirrors your text selection - If you searched multiple alternatives or categories, include selected URLs from all relevant searches in this one call - 1–10 items; never pass more than you actually recommend in text What happens The widget renders only these listings as product cards. smart_search and compare auto-display their server-selected recommendations when a single search returns relevant results; use this tool only when you need to replace that automatic display with your curated subset. Rules - Always use product_url values exactly as returned by the search tool — never rewrite or reconstruct URLs - This tool can only render URLs returned by recent search tools. If you do not have exact current product_url values, search first. - If the user switched language since the last search, run a new search with the new locale first; old-locale URLs are not valid for the new locale - This is always the last tool call before your text response when results are found - Never call this for an empty result, weak match, exploratory search, diagnostic candidate set, or a result you plan to replace with another search Args - product_urls: ordered list of product_url strings from the final selected recommendations. Best match first.
display_recommendations
ChatGPTSend a curated selection to the Tawadoo widget. Search tools are data-only and do not render the widget. Call this tool exactly once after a final relevant search result, using the exact product_url values of the listings you will recommend in text. The widget will display those listings as product cards — nothing more, nothing less. When to call - A search returned relevant recommendations and you have picked your 1–10 final recommendations → call this exactly once before answering - Multiple relevant searches were needed and you have picked your 1–10 final recommendations → call this exactly once before answering - One broad search returned many relevant listings and you want the widget to show only your final subset → call this exactly once before answering - is_relevant=false or no recommendations → do not call this (nothing to display) What to pass - product_urls: the ordered list of product_url values of the listings you are recommending, best match first - Pass exactly the URLs you will link in your text response — the widget mirrors your text selection - If you searched multiple alternatives or categories, include selected URLs from all relevant searches in this one call - 1–10 items; never pass more than you actually recommend in text What happens The widget renders only these listings as product cards. smart_search and compare auto-display their server-selected recommendations when a single search returns relevant results; use this tool only when you need to replace that automatic display with your curated subset. Rules - Always use product_url values exactly as returned by the search tool — never rewrite or reconstruct URLs - This tool can only render URLs returned by recent search tools. If you do not have exact current product_url values, search first. - If the user switched language since the last search, run a new search with the new locale first; old-locale URLs are not valid for the new locale - This is always the last tool call before your text response when results are found - Never call this for an empty result, weak match, exploratory search, diagnostic candidate set, or a result you plan to replace with another search Args - product_urls: ordered list of product_url strings from the final selected recommendations. Best match first.
get_buyer_guidance
ChatGPTUse this tool when the user is a buyer asking any question about the Tawadoo platform, its rules, or how to buy safely and effectively. Call it for questions about: how to search for a product on Tawadoo, how to use filters or search by image, the difference between Fixed Price / Negotiable / Live Auction listings and how to interact with each, how to make an offer on a Negotiable listing and what strategy to use, how to participate in a live auction, what Tawadoo Coins are and how to get them as a buyer, how to evaluate whether a seller is trustworthy (Verified badge, ratings, reviews), how payment escrow works and why it protects buyers, what payment methods are accepted, how delivery works via Tawssil.ma, what Relay Points (Points Relais) are and how to use them, how to track a delivery, what buyer protection covers and how to open a dispute, what to do if an item is not as described or not received, how to report a suspicious listing or seller, how the review system works, how to save favourite listings and receive price-drop alerts, or any other platform process question from a buyer's perspective. Do NOT call this tool for product search or listing discovery requests — use the dedicated search tools for those.
get_buyer_guidance
ChatGPTUse this tool when the user is a buyer asking any question about the Tawadoo platform, its rules, or how to buy safely and effectively. Call it for questions about: how to search for a product on Tawadoo, how to use filters or search by image, the difference between Fixed Price / Negotiable / Live Auction listings and how to interact with each, how to make an offer on a Negotiable listing and what strategy to use, how to participate in a live auction, what Tawadoo Coins are and how to get them as a buyer, how to evaluate whether a seller is trustworthy (Verified badge, ratings, reviews), how payment escrow works and why it protects buyers, what payment methods are accepted, how delivery works via Tawssil.ma, what Relay Points (Points Relais) are and how to use them, how to track a delivery, what buyer protection covers and how to open a dispute, what to do if an item is not as described or not received, how to report a suspicious listing or seller, how the review system works, how to save favourite listings and receive price-drop alerts, or any other platform process question from a buyer's perspective. Do NOT call this tool for product search or listing discovery requests — use the dedicated search tools for those.
get_seller_guidance
ChatGPTUse this tool when the user is a seller asking any question about the Tawadoo platform, its rules, or how to sell effectively. Call it for questions about: how to create a listing on Tawadoo, which category to choose for a product, which selling mode fits best (Fixed Price vs Negotiable vs Live Auction), how to set a competitive price, how to write a good title or description, how photos and videos affect performance, how Tawadoo Coins work, how to earn free Coins, what the Lucky Wheel is, how Visibility Boosts work and when to use them, what DaaS subscription plans offer and which plan to choose, how live auctions work (anti-sniping, Coin escrow, non-paying winner handling), what the transaction fees are, how the seller ranking algorithm works, why response speed matters, how to read seller analytics, how to get the Verified Seller badge, how payments and delivery work with Tawssil.ma, how to edit or pause a listing, what items are prohibited, or any other platform process question from a seller's perspective. Do NOT call this tool for product search or listing discovery requests — use the dedicated search tools for those.
get_seller_guidance
ChatGPTUse this tool when the user is a seller asking any question about the Tawadoo platform, its rules, or how to sell effectively. Call it for questions about: how to create a listing on Tawadoo, which category to choose for a product, which selling mode fits best (Fixed Price vs Negotiable vs Live Auction), how to set a competitive price, how to write a good title or description, how photos and videos affect performance, how Tawadoo Coins work, how to earn free Coins, what the Lucky Wheel is, how Visibility Boosts work and when to use them, what DaaS subscription plans offer and which plan to choose, how live auctions work (anti-sniping, Coin escrow, non-paying winner handling), what the transaction fees are, how the seller ranking algorithm works, why response speed matters, how to read seller analytics, how to get the Verified Seller badge, how payments and delivery work with Tawssil.ma, how to edit or pause a listing, what items are prohibited, or any other platform process question from a seller's perspective. Do NOT call this tool for product search or listing discovery requests — use the dedicated search tools for those.
search_family_pets
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace baby/mom and pet listings using static branches generated from the current taxonomy. Sections: all, toys_games, baby_room, baby_development, baby_clothes, baby_feeding, baby_safety, maternity_nursing, baby_outings_travel, strollers, car_seats, baby_hygiene, pet_food, pet_bedding, pet_accessories, pet_training, pet_travel_walk. search_text must be the item type (e.g. 'poussette', 'siège auto', 'jouet') — never brand names. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_family_pets
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace baby/mom and pet listings using static branches generated from the current taxonomy. Sections: all, toys_games, baby_room, baby_development, baby_clothes, baby_feeding, baby_safety, maternity_nursing, baby_outings_travel, strollers, car_seats, baby_hygiene, pet_food, pet_bedding, pet_accessories, pet_training, pet_travel_walk. search_text must be the item type (e.g. 'poussette', 'siège auto', 'jouet') — never brand names. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_fashion_beauty
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace fashion/beauty listings using static branches generated from the current taxonomy. Sections: all, clothing, shoes, jewelry, luggage_leather_goods, accessories, makeup, perfumes, care, health, watches, glasses. search_text must be the item type or style (e.g. 'robe', 'sneakers', 'montre', 'sac à main') — never brand names. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_fashion_beauty
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace fashion/beauty listings using static branches generated from the current taxonomy. Sections: all, clothing, shoes, jewelry, luggage_leather_goods, accessories, makeup, perfumes, care, health, watches, glasses. search_text must be the item type or style (e.g. 'robe', 'sneakers', 'montre', 'sac à main') — never brand names. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_high_tech
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace high-tech listings using static branches generated from the current taxonomy. Sections: all, smartphones, tablets, connected_devices, phone_accessories, computers, accessories_peripherals, gaming, printers_scanners, tv_home_cinema, audio_music, photo_camera. search_text must be the product type or key spec in the user's locale/listing vocabulary (e.g. 'smartphone', 'ordinateur portable', 'tablette 10 pouces', 'écran 4K') — never brand names or model references, and add qualifiers like 'gaming' only when the user explicitly asks for them. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_high_tech
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace high-tech listings using static branches generated from the current taxonomy. Sections: all, smartphones, tablets, connected_devices, phone_accessories, computers, accessories_peripherals, gaming, printers_scanners, tv_home_cinema, audio_music, photo_camera. search_text must be the product type or key spec in the user's locale/listing vocabulary (e.g. 'smartphone', 'ordinateur portable', 'tablette 10 pouces', 'écran 4K') — never brand names or model references, and add qualifiers like 'gaming' only when the user explicitly asks for them. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_home_living
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace home/living/appliances/DIY listings using static branches generated from the current taxonomy. Sections: all, furniture, kitchen, decoration, storage_organization, home_linen, lighting, appliances, kitchen_cooking_appliances, maintenance_appliances, heating_air_conditioning, tools_hardware, security_home_automation, home_staging, plumbing_heating, garden_outdoor, household_products. search_text must be the item type (e.g. 'canapé', 'machine à laver', 'climatiseur', 'réfrigérateur') — never brand names. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_home_living
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace home/living/appliances/DIY listings using static branches generated from the current taxonomy. Sections: all, furniture, kitchen, decoration, storage_organization, home_linen, lighting, appliances, kitchen_cooking_appliances, maintenance_appliances, heating_air_conditioning, tools_hardware, security_home_automation, home_staging, plumbing_heating, garden_outdoor, household_products. search_text must be the item type (e.g. 'canapé', 'machine à laver', 'climatiseur', 'réfrigérateur') — never brand names. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_real_estate
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace real-estate listings using static branches generated from the current taxonomy. Sections: all, for_sale, for_rent, apartments_for_sale, houses_for_sale, land_for_sale, commercial_for_sale, apartments_for_rent, houses_for_rent, rooms_for_rent, commercial_for_rent. search_text must be the property type (e.g. 'appartement', 'villa', 'studio', 'terrain', 'bureau') — never city names, district names, or user context words. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_real_estate
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace real-estate listings using static branches generated from the current taxonomy. Sections: all, for_sale, for_rent, apartments_for_sale, houses_for_sale, land_for_sale, commercial_for_sale, apartments_for_rent, houses_for_rent, rooms_for_rent, commercial_for_rent. search_text must be the property type (e.g. 'appartement', 'villa', 'studio', 'terrain', 'bureau') — never city names, district names, or user context words. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_services_jobs
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace services and job listings using static branches generated from the current taxonomy. Sections: all, auto_services, consultancy, domestic_services, event_entertainment, freelancers, health_wellbeing, home_maintenance, movers_removals, restoration_repairs, tutors_classes, web_computer_services, other_services, job_offers, job_requests. search_text must be the service or job type (e.g. 'plombier', 'cours anglais', 'développeur web') — never company names or user context words. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_services_jobs
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace services and job listings using static branches generated from the current taxonomy. Sections: all, auto_services, consultancy, domestic_services, event_entertainment, freelancers, health_wellbeing, home_maintenance, movers_removals, restoration_repairs, tutors_classes, web_computer_services, other_services, job_offers, job_requests. search_text must be the service or job type (e.g. 'plombier', 'cours anglais', 'développeur web') — never company names or user context words. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_vacation_rentals
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace vacation rental listings using static branches generated from the current taxonomy. Sections: all, hotels, seasonal_rentals, wellness, corporate_travel. search_text must be the accommodation type (e.g. 'riad', 'villa vacances', 'appartement meublé') — never city, region, or date references. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_vacation_rentals
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace vacation rental listings using static branches generated from the current taxonomy. Sections: all, hotels, seasonal_rentals, wellness, corporate_travel. search_text must be the accommodation type (e.g. 'riad', 'villa vacances', 'appartement meublé') — never city, region, or date references. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_vehicles
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace vehicle listings using static branches generated from the current taxonomy. Sections: all, cars, motorcycles, commercial_vehicles, boats, bikes, parts_accessories. search_text must be the vehicle type or use case (e.g. 'citadine', 'SUV', 'berline', 'utilitaire', '4x4') — never brand names, model references, or context words from the user's message. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
search_vehicles
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search marketplace vehicle listings using static branches generated from the current taxonomy. Sections: all, cars, motorcycles, commercial_vehicles, boats, bikes, parts_accessories. search_text must be the vehicle type or use case (e.g. 'citadine', 'SUV', 'berline', 'utilitaire', '4x4') — never brand names, model references, or context words from the user's message. search_text rule: Leave search_text EMPTY when the user's request is just the category or section itself — e.g. 'montre-moi des smartphones', 'je cherche une voiture', 'appartements à louer' — because the section filter already covers this and adding the category word as search_text would only exclude listings that don't mention it literally. Only populate search_text when the user specifies something more specific than the section: a sub-type, style, feature, size, or material (e.g. 'cabriolet', 'studio', 'gaming', 'cuir', '4x4'). This applies regardless of language or phrasing — 'téléphone', 'telephone', 'هاتف', 'gsm' are all just the Smartphones section and search_text should stay empty. Put brand/model/spec qualifiers in ranking_query only, keeping it short. For alternative brands or models, separate with '|', e.g. 'Audi|Mercedes', in one call.
smart_search
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search Tawadoo in one call — automatic category resolution + product search. Use this when the user gives a broad or ambiguous request and no static vertical tool clearly applies. The server resolves the best category within a bounded search budget, uses cacheable taxonomy metadata, and returns ranked results ready to evaluate. This search tool is data-only. It does not render the widget. When recommendations are available and this is the final relevant search, call display_recommendations exactly once with the product_url values you will mention in text. smart_search is generic. It does not contain language-specific synonym tables or category-specific business rules. You are responsible for interpreting the user's natural language before calling. For marketplace listing requests, never use web search or outside market knowledge. If the request maps to Tawadoo, use Tawadoo MCP tools only. Do not use smart_search for a completely categoryless recommendation request such as "what do you recommend on the site?" when the user gives no product type, use case, budget, or preference. Ask one short clarifying question instead of choosing a category. When to use | Situation | Use | |---|---| | Request clearly maps to a major vertical | Use the static vertical tool directly | | Broad, ambiguous, or cross-category request | Use smart_search | | Structured criteria + clear vertical | Use the vertical tool — do not default to smart_search | If a brand/model is paired with an implied product vertical, use the static vertical tool rather than smart_search. smart_search is for unresolved category discovery; it can legitimately surface cross-category matches when the query is only a brand/model with no product type. If smart_search resolves the wrong category branch, correct silently with the appropriate vertical tool. Never tell the user that smart_search confused categories or resolved the wrong branch. If the exact same request with the same locale and filters was already answered in this conversation, reuse that result instead of calling again. Re-call only when locale, budget, filters, or page changed. Before calling — decompose the request - Hard criteria (brand, model, capacity, city, price, fuel type, rooms): keep these in query or pass as params; do not drop them silently - Soft criteria (condition words: new, like new, clean, no scratches): use a broad query for recall; explain the condition caveat in your answer - Broad recommendation wording ("best value", "rapport qualité prix", "pas cher mais bien"): search for the core object (query="smartphone"), not the recommendation words; compare returned listings by price, condition, and negotiability `query` must be short and verbatim-safe. Before including any word, apply this test: "Would this word appear in a listing title or structured property field?" If no, omit it. Keep query to 1–4 terms. - User context (situation, usage, lifestyle) → interpret it, then drop it from query - Quality adjectives (fiable, économique, propre, bon état) → omit; they appear in some descriptions but not all, and their absence lowers the relevance score unfairly - Budget amounts → pass as the budget param, never in query For broad category searches (user wants "a car", "a phone" with few hard constraints): use 1–2 core product nouns only. A long query inflates query_term_count and can push the match ratio below the relevance threshold even when matching listings exist. If is_relevant=false, check whether query was too long or contained words unlikely to appear verbatim in listings. If so, retry once with a shorter, core-noun-only query. For storage or unit wording, normalize before calling…
smart_search
ChatGPTDefault Tawadoo marketplace tool. Use it whenever the user asks to find, buy, browse, compare, recommend, or show real marketplace listings, with or without budgets or preferences, even when the user does not explicitly say Tawadoo. Do not use web search, browser search, or outside market knowledge for matching listing requests. Search Tawadoo in one call — automatic category resolution + product search. Use this when the user gives a broad or ambiguous request and no static vertical tool clearly applies. The server resolves the best category within a bounded search budget, uses cacheable taxonomy metadata, and returns ranked results ready to evaluate. This search tool is data-only. It does not render the widget. When recommendations are available and this is the final relevant search, call display_recommendations exactly once with the product_url values you will mention in text. smart_search is generic. It does not contain language-specific synonym tables or category-specific business rules. You are responsible for interpreting the user's natural language before calling. For marketplace listing requests, never use web search or outside market knowledge. If the request maps to Tawadoo, use Tawadoo MCP tools only. Do not use smart_search for a completely categoryless recommendation request such as "what do you recommend on the site?" when the user gives no product type, use case, budget, or preference. Ask one short clarifying question instead of choosing a category. When to use | Situation | Use | |---|---| | Request clearly maps to a major vertical | Use the static vertical tool directly | | Broad, ambiguous, or cross-category request | Use smart_search | | Structured criteria + clear vertical | Use the vertical tool — do not default to smart_search | If a brand/model is paired with an implied product vertical, use the static vertical tool rather than smart_search. smart_search is for unresolved category discovery; it can legitimately surface cross-category matches when the query is only a brand/model with no product type. If smart_search resolves the wrong category branch, correct silently with the appropriate vertical tool. Never tell the user that smart_search confused categories or resolved the wrong branch. If the exact same request with the same locale and filters was already answered in this conversation, reuse that result instead of calling again. Re-call only when locale, budget, filters, or page changed. Before calling — decompose the request - Hard criteria (brand, model, capacity, city, price, fuel type, rooms): keep these in query or pass as params; do not drop them silently - Soft criteria (condition words: new, like new, clean, no scratches): use a broad query for recall; explain the condition caveat in your answer - Broad recommendation wording ("best value", "rapport qualité prix", "pas cher mais bien"): search for the core object (query="smartphone"), not the recommendation words; compare returned listings by price, condition, and negotiability `query` must be short and verbatim-safe. Before including any word, apply this test: "Would this word appear in a listing title or structured property field?" If no, omit it. Keep query to 1–4 terms. - User context (situation, usage, lifestyle) → interpret it, then drop it from query - Quality adjectives (fiable, économique, propre, bon état) → omit; they appear in some descriptions but not all, and their absence lowers the relevance score unfairly - Budget amounts → pass as the budget param, never in query For broad category searches (user wants "a car", "a phone" with few hard constraints): use 1–2 core product nouns only. A long query inflates query_term_count and can push the match ratio below the relevance threshold even when matching listings exist. If is_relevant=false, check whether query was too long or contained words unlikely to appear verbatim in listings. If so, retry once with a shorter, core-noun-only query. For storage or unit wording, normalize before calling…
