add_to_wishlist
ChatGPTAdd an auction to the user's wishlist. Idempotent. Auto-creates a reminder (1 day before, falling back to 1 hour).
analyze_auction
ChatGPTRecent nearby auction results (not market sales), area auction statistics, risk factors, and LTR/BRRRR/Flip/STR deal metrics for one auction, from the same engine as the AuctionScout map pins. deal gives each strategy's status (a reason when it is pending), the discount to market, the inputs used and warnings. Numbers only, no AI narrative. deal_metrics.brrrr_cash_on_cash of 999 means all cash is recovered at refinance (deal.brrrr.all_cash_out is true).
create_reminder
ChatGPTCreate a reminder for an auction. minutes_before must be 15, 60, or 1440.
create_saved_search
ChatGPTSave a named search with filters. filters is a predictions-filter object — see the schema for the exact fields. Range filters such as market_price_range are 2-element [min, max] arrays (e.g. [0, 500000]), not {min, max} objects. Daily new-match alerts (email + push) are on by default; pass alerts_enabled=false to opt out. Alert frequency is configured in the portal. Check the result's ignored_keys and ignored_values: anything listed there was NOT saved, so tell the user.
delete_reminder
ChatGPTDelete one of the user's auction reminders. Get the reminder_id from list_reminders (the id field) and confirm with the user before deleting. Returns deleted=false if no such auction reminder exists for this user.
get_auction
ChatGPTFull detail for one auction, including sold_price for completed sales, parcel id, county auditor/clerk URLs, and a (DB-only) estimates bundle with the market value (purchase_price: the resale value, not the buy price) and the model's predicted winning bid (predicted_price). Estimate fields may be null.
get_neighborhood
ChatGPTCached neighborhood data for an auction (looked up by auction_id). Sections: schools (district name, PI rating), amenities (nearby grocery/dining/parks/etc. and walkability score), demographics (census tract income/population/home values). Returns null for any section not yet in cache — no live API calls are made. If the school district isn't cached, schools instead falls back to the auction record's nearby-school rating (source: auction_record). Each missing or fallback section is explained in missing_reasons (no_coordinates, not_cached_yet, cache_expired, read_error). missing_reasons, not partial, is the complete signal: partial stays false when schools comes from the fallback.
get_predictions
ChatGPTUpcoming auctions with model-predicted sale prices and confidence scores. Auctions without a prediction row are included with prediction=null. Filter by state, county, min_confidence, or max_days_ahead. Up to 50 rows; total_count is the full match count, and a coverage object (same as search_auctions) says when a state isn't covered. ROI = (market_value - predicted_price) / predicted_price * 100, where market_value is the same market estimate search_auctions and get_auction return (latest market price estimate, else the automated home-value estimate). Cancelled, withdrawn, dismissed, vacated, redeemed and already-decided auctions (sold, no sale, etc.) are excluded; postponed and bankruptcy rows stay, labeled by sale_status. Each row carries state and sale_status. is_fallback=true means the price is a formula (market estimate times a historical range edge, or a $1,000 / $10M safety limit), not a model estimate; it carries low confidence and a fallback_reason. market_checked=false means the model price was made before the property had a market estimate, so nothing checked it (null for older rows).
list_reminders
ChatGPTThe user's auction reminders. Call with NO arguments to list every reminder they have; pass auction_id to list only that auction's reminders. Each row includes the auction address and date.
list_saved_searches
ChatGPTReturn the authenticated user's saved searches.
list_wishlist
ChatGPTReturn the user's wishlisted auctions. filter: 'upcoming' (default) | 'past' | 'all'.
remove_from_wishlist
ChatGPTRemove an auction from the user's wishlist. Idempotent (no error if absent). Also deletes the unsent reminder that add_to_wishlist created automatically; reminders the user set themselves are kept.
search_auctions
ChatGPTSearch auctions by state, county, city, date window, value range, sale_status, or address. Use this for any 'upcoming auctions' or 'auctions in <place>' request. With no date_from/date_to it returns only upcoming auctions (today onward, US/Eastern); upcoming hides sales that are already decided (sold, no sale, returned to lender) and keeps postponed and bankruptcy rows, labeled by sale_status. Cancelled, withdrawn, dismissed, vacated and redeemed sales are hidden by default; pass an explicit sale_status to include them. Set state whenever the user names a state, even one we may not cover: the response always includes a coverage object that says when the state or county isn't covered and names what we do serve. Returns 20 rows by default (limit up to 50) plus total_count; when next_offset is set, call again with offset=next_offset for the next page. Upcoming searches list the soonest first; date_to-only or past searches list the newest first (override with sort). Value filters use the appraisal when there is one, else the market estimate (each row's value_basis says which). Each row has a portal_url deep link; the response echoes filter_applied. Sold rows carry sold_price (the winning bid; null for returned-to-lender credit bids).