add_note
ChatGPTAdd a note to a lead — the same note the pipeline card's Notes sheet writes, visible to the user's team when the lead is on a team board. Useful for capturing call outcomes, quote details, follow-up reasons. There is no undo from chat; the user can delete it on the card.
add_to_watchlist
ChatGPTAdd a builder COMPANY to the user's watchlist so its new permits surface first and score higher. Use it for every way of saying so: 'watch Polygon', 'add them to my watchlist', 'follow Strand', 'track Adera', 'keep an eye on Qualico', 'tell me when Mosaic pulls a permit'. 'Follow up with X' is NOT a watch, and a specific job to put on the board is create_lead; if unclear, ask one short question. companyId is a companies-table uuid (the builder's id from a get_permit_detail / query_permits result) — NOT a permit id. Confirm in your reply; the change takes effect on the user's next page load.
coverage_check
ChatGPTFree coverage check for an area, no sign-in needed: how many permits BuildMapper holds there, how many builders are actively permitting, the estimated addressable supply spend, and the builders filing the most work. Read from BuildMapper's shared coverage templates for that area. Pass cities, or regions, or a province.
create_contract
ChatGPTStart a draft contract in BuildMapper's Contracts app from a pipeline lead, a permit, or an existing quote (the quote's client, value and line items carry over). Returns the contract's id and the link where the user finishes and sends it. This never sends anything.
create_lead
ChatGPTCreate a NEW manual lead in the user's pipeline for a project that is NOT one of the permits in the database — a referral, a phone tip, a project the user names. (For a lead that DOES come from a permit you surfaced, the user saves it from the project drawer; don't use this for those.) Confirm what you created in your reply, and offer to undo (dismiss_lead) if they didn't want it.
create_quote
ChatGPTStart a draft quote in BuildMapper's Quotes app, optionally prefilled from a pipeline lead or a permit. Returns the quote's id and the link where the user finishes and sends it. This never sends anything.
delete_calendar_event
ChatGPTRemove ONE of the user's calendar events by id (from list_calendar or a schedule_activity result). The undo for schedule_activity. Confirm what you removed in your reply.
dismiss_lead
ChatGPTSoft-delete a lead from the user's pipeline. Use when the user explicitly says to drop / dismiss / archive a lead.
draft_email
ChatGPTCompose a ready-to-send outreach email draft for ONE project (permit). Grounds the email in the user's trade + saved email signature and the project's real builder/contact — the SAME drafter the project page's 'Draft with AI' uses. Returns the finished draft {to, subject, body}. Present the returned draft to the user verbatim — do not rewrite the subject or body. To draft for several projects at once, call this once per project (up to ~4 per turn to stay within the reply limit) and present each draft. The user reviews and sends every draft themselves from their own mail app; you NEVER send email.
erp_account_summary
ChatGPTFull picture of one ERP account: identity, rep, trailing-12-month vs prior-12-month revenue, first/last order dates, open orders, open quotes with value, AR balance, plus its most recent orders and quotes. Pass the account name or its ERP customer number.
erp_dormant_accounts
ChatGPTAccounts that went quiet: real customers (they have order history) whose LAST order is older than a threshold, ranked by how much they used to buy. The win-back list — 'who should I call that we're losing?'. Defaults: no order in 90 days.
erp_find_account
ChatGPTSearch the connected ERP's customer accounts by name (also matches city and assigned rep). Returns compact rows with revenue and activity rollups. Use this first to resolve 'the account' a user mentions before deeper lookups.
erp_recent_orders
ChatGPTList recent order or quote headers across ALL ERP accounts — 'what's on order this week', 'open quotes right now', 'what did we close last month'. Filter by kind (order/quote), status, rep, and a day window. For one specific account use erp_account_summary instead.
find_by_contact
ChatGPTLook up permits by a PHONE NUMBER or a PERSON'S NAME (someone's first/last name — for a COMPANY use query_permits' search). Reads the permit contacts the user is allowed to see, builder office lines visible to them, and their own saved Connections contacts, and returns the matching people plus the permits they are attached to (an office line → that builder's permits from the last 18 months). Always call this before saying a number or a name can't be found; when matches is empty say so plainly and offer a web search. Locked projects come back with their area-only address.
get_company_intel
ChatGPTWhat saved research knows about a COMPANY (a builder, developer or owner): profile, scale, what and where they build, posture toward subcontractors, memberships, recent projects, website / LinkedIn / phone / email, suggested approach, sources, and the principals found at the firm with a confidence each — plus how many permits of theirs are on the map. Use for "tell me about Acme Homes", an account brief, or before Find phone / Find email (a blank field means nothing is saved for it yet). companyId from a permit result or the hidden context; or a name (+ city) when that is all you have — the result says whether the match was exact.
get_permit_detail
ChatGPTFetch full details for one permit by id, including the builder company and its contacts. Use when the user asks about a specific permit you already surfaced.
get_project_intel
ChatGPTEverything already KNOWN about a project, read from saved research — this permit's own project research (if anyone ran it), the profile of its builder and owner learned on ANY of their permits (summary, scale, what they build, posture toward subcontractors, website / phone / email, recent projects, suggested approach, sources, and which address it was learned on), and the people found at those firms with a confidence each. Check it before saying nothing is known. permitId from a permit result or the hidden context. stale means the research is old — say when it was researched. Empty companies and no project means nobody has researched this builder yet; the Research card on the project page runs it.
get_project_research
ChatGPTThe RAW research stages saved on one project (project, companies, people), if any — the full card text. Prefer get_project_intel, which adds what is known about the builder from its other permits and is shaped to answer from; use this only when the user asks for the full write-up of this exact project. permitId from a permit result or the hidden context.
get_takeoff
ChatGPTRead ONE of the user's finished plan takeoffs in full: every material line (section, name, qty, unit, counted qty and waste % when their waste rule is on, price from their own price list or needs_pricing, the sheets it was read from, how it was measured), the flags raised while reading, the rules it used, the next step, and the review link. Pass takeoff_id from list_my_takeoffs, or job words to find it. Read-only: to change a line, price it or put it on a quote, send them to the review link.
import_customer_list
ChatGPTSave a list of companies as the user's customers in BuildMapper — the explicit follow-up to match_companies, only when the user asks for the list to be kept. Writes through the same two paths a dropped file uses: for a team owner or manager whose branch has opened Territory Intelligence, the list becomes a named-accounts file in their territory workspace (it shows under Settings → Your data with a receipt and can be removed there); for everyone, each company (and any person named on it) becomes a Connections contact matched to its builder. Duplicates already on the account are skipped, so re-running is safe. A rep's or solo user's matched builders are watched; a manager's are not (their file is the branch's book). Up to 500 rows per call.
integration_status
ChatGPTWhat this BuildMapper account has connected and given us: ERP and CRM connections, Product Brain (the catalog), the Territory workspace and its customer files, Connections contacts, the account's own permit files, and watched builders — with where each is set up in the app. Call it before suggesting an import or when the user asks what BuildMapper already knows about their customers, catalog or systems.
list_calendar
ChatGPTThe user's upcoming calendar events (site visits, calls, follow-ups, deadlines) for the next N days, soonest first, with the event id (needed for delete_calendar_event) and the linked project. Read this for 'what's on this week', 'when am I seeing X', 'did I book that'.
list_categories
ChatGPTThe catalog's category and material-type taxonomy with product counts. Cheap — call it before filtering by category so you use the team's actual labels.
list_documents
ChatGPTList the files in the knowledge base with their type, status and summary. Use for 'what do we have on X', 'did the new price list get uploaded', or to check source freshness.
list_inbox
ChatGPTThe user's open alerts (the bell): new permits from watched builders, hot leads, stale pipeline, niche matches, quote viewed, weekly digest. Newest first, with the project each points at. Read this for 'what's new', 'anything I missed', 'watchlist activity'. Read-only; the user clears them in the bell.
list_lead_notes
ChatGPTThe notes on one lead (the pipeline card's Notes sheet), newest first. leadId from list_my_pipeline. Read this before summarising a lead's history or 'what did I say about X'.
list_my_documents
ChatGPTThe user's quotes, contracts and invoices with status, total, client, when it was sent and whether the client has VIEWED it (viewed_at) or responded / signed / paid. Newest first. Filter by kind, status, lead or permit. Read this for 'did they open my quote', 'what's outstanding', 'unpaid invoices'. The user's own documents only.
list_my_pipeline
ChatGPTList the user's saved leads, optionally filtered by stage. Returns each lead's id, stage, score, last contact, quoted/won value, next action and the joined permit's address + builder (or the manual lead's own title/address/builder). When the user's team has custom pipeline columns, stages lists them (id + label) — use those ids with move_lead_stage.
list_my_takeoffs
ChatGPTList the user's plan takeoffs (Quotes & Takeoffs → Takeoffs), newest first: name, status, how many materials, how many still need a price, the quote it is on, and a link to review it beside the drawings. Read-only.
list_team_pipeline
ChatGPTTEAM OWNERS AND MANAGERS ONLY (see the user's role in the context block): the whole team's pipeline, per rep — how many leads each rep has by stage and their most recent open leads — or one rep's board with repUserId. The same rows the pipeline's rep switcher shows. Refuses for a rep. Never move or reassign another rep's lead from here; assignment goes through the Team app.
list_watchlist
ChatGPTThe builders the user is watching, with how many permits each filed in the last 90 days and the newest one. Read this for 'who am I watching', 'any activity from my watched builders', 'am I already watching X'.
match_companies
ChatGPTCheck a list of company names against BuildMapper's permit database WITHOUT importing anything: which of them are builders or owners on file, and what each has filed recently, scored for the user. Use it on names read from whatever the user has connected to the assistant — a spreadsheet, a CRM, an ERP export, an email — to answer "which of my customers are building right now" or "who on this list has a permit in". Returns one row per matched company (confidence, method, whether it is an inferred guess below the confirm threshold, permits in the last 12 months) with its newest permits inside the window, plus the names that matched nothing. Up to 200 names per call; send the rest in another call. Nothing is stored — offer import_customer_list afterwards if the user wants the list kept as their customers.
move_lead_stage
ChatGPTMove a lead between pipeline stages. Pass the lead id (NOT the permit id) and the new stage: 'new', 'contacted', 'quoted', 'won', 'lost', or a custom column id from list_my_pipeline's stages. When the user names an amount, pass it too: contractValue for a quote ('quoted them 15k'), valueWon for a closed deal. Returns error 'lead_not_found' if that lead isn't on the user's board — when you get that, do NOT claim you moved it; the project isn't saved to the pipeline yet, so save it (or create_lead) first, then move it. 'unknown_stage' returns the columns that exist — pick one, never invent an id. Confirm the action in your reply.
plan_route
ChatGPTOrder a set of projects into the shortest drive and return a Google Maps directions link (one link per 10 stops). Pass permitIds (from query_permits / list_my_pipeline results) or leadIds, and where the user starts: startLat/startLng, or startAddress (geocoded), or nothing to start from the first stop. Use it for 'plan my day', 'route these', 'what order should I visit', 'I'm at X, where first'. Up to 20 stops. Locked projects are placed by their coordinates but their address stays area-only.
query_permits
ChatGPTSearch the permit database by place, type, status, value, score, date or free text (description, address, BUILDER/COMPANY name). Use this for any 'find / show me / how many' question about permits. Filters are AND-combined; arrays are OR-within-field. For a PHONE NUMBER or a PERSON's name use find_by_contact instead — this search does not read contacts. hitCount is capped at the sample ceiling: when capped is true say 'more than N', never an exact total.
read_guide
ChatGPTRead one Learn guide in full by its slug (the slugs are listed under 'Learn library' in your instructions). Use it whenever a support answer needs more than the product guide's one-liners: an exact setup sequence, what a page or tab shows, a policy (data, billing, cancellation), or a playbook. Answer from the text, keep it to the steps the user asked for, and point them at the guide's link for the rest.
remove_from_watchlist
ChatGPTRemove a builder company from the user's watchlist. companyId is the companies-table uuid (NOT a permit id).
save_company_contact
ChatGPTPersist contact info found for a builder company. Use this only when ALL of: (a) the source is the company's own website/domain or a LinkedIn page that links back to it, (b) the company name on the source matches the target company, (c) the source's location (city/state) overlaps with the company's known territory or one of its permit cities. Otherwise show the finding to the user to confirm in BuildMapper instead of saving it. companyId comes from the seeded prompt or a get_permit_detail / query_permits result.
save_permit_as_lead
ChatGPTSave a permit (a project from query_permits / get_permit_detail) to the user's pipeline as a new lead in the 'new' stage, with their live score. Idempotent: a project already in the pipeline is reported, not duplicated. For a lead that is NOT a permit (an off-permit opportunity) use create_lead instead.
schedule_activity
ChatGPTAdd an event to the user's calendar — a call, site visit, follow-up, or deadline. Provide ISO-8601 times. If the user gives only a day (no time), set allDay:true. Optionally tie it to a permit. State the date/time back in your reply so they can confirm.
search_guides
ChatGPTKeyword search over BuildMapper's Learn library — the product's own how-to guides, playbooks, glossary and FAQ. Use it for a how-do-I / support / policy question when the guide slug isn't obvious from the index in your instructions. Returns the best-matching guides with a snippet each; call read_guide on the best one when the snippet isn't enough.
search_knowledge
ChatGPTSemantic search across everything in the team's uploaded documents (spec sheets, flyers, catalogs, price lists). Use for specs, warranties, installation guidance, ratings, promo terms — any question that isn't a pure price/SKU lookup. Results include the source filename and page: cite them. NEVER quote an exact price from these passages — confirm prices via search_products.
search_products
ChatGPTSearch the team's structured product catalog. THE ONLY SOURCE FOR EXACT PRICES, SKUs and units. Text search matches name/SKU/description, ranked by relevance with an exact SKU always first; a semantic pass fills in so fuzzy names still resolve. Products with status 'needs_review' have unverified extracted data — say so when quoting them. price_effective_date is the vintage of the price book the price came from (null = the document didn't state one).
set_next_action
ChatGPTSet the recommended next action on a lead — it shows on the pipeline card. leadId is the lead's id (NOT a permit id; use list_my_pipeline to find it). Confirm what you set in your reply.
territory_get_assumptions
ChatGPTThe published assumptions behind the numbers: the product-line fit weights, the blended spend share, and the minimum permits before a builder counts as actively permitting. Use when the manager disputes a figure - usually the answer is an assumption they can correct, not a bug.
territory_get_builder_permits
ChatGPTThe actual permits for one builder brand. Use when the manager asks about a specific builder, or to show why a gap-list entry is worth their time.
territory_get_covered_accounts
ChatGPTThe manager's named accounts that matched a builder, with match confidence and method. Use when asked what they already cover, or to point out matches that were only inferred and should be checked.
territory_get_gap_list
ChatGPTBuilders active in the territory that the manager has NOT named, ranked by estimated addressable spend. Use for 'who should we go after' questions.
territory_get_market_context
ChatGPTWeb research about the manager's own company and other suppliers in their region. THIS IS NOT PERMIT DATA - it is web research with sources, and it may be missing or stale. Always say where it came from when you use it, and never mix it into a coverage or opportunity figure.
territory_get_summary
ChatGPTThe headline coverage picture: permits in territory, relevant permits, how many builders are actively permitting, how many of the manager's named accounts matched, and total estimated addressable spend. Call this first when asked anything about the overall numbers.
territory_get_team_coverage
ChatGPTWhere each rep's saved pipeline sits across the territory, which zones have real opportunity but no open leads, and the closest unnamed builders to each rep's existing patch. Manager-only; returns a reason instead of data if the caller is a rep or has no team. CRITICAL: this reflects pipeline SAVED IN BUILDMAPPER only. A rep working an account by phone without saving it is invisible here. Never say a rep is neglecting an area or underperforming - say there is no saved pipeline there, and suggest the manager ask. Never assign work to a named rep.
territory_get_zone_permits
ChatGPTThe actual permits behind a zone. Use to ground a claim in specific projects, or when the manager asks to see what is in an area.
territory_get_zones
ChatGPTEvery zone with its opportunity score, recommendation (expand/maintain/reduce/senior), builder counts, coverage, trend, and the SCORE COMPONENT BREAKDOWN. Use this to explain why a zone scored the way it did - the components carry the real contributing factors, so never guess at a rationale.
undo_action
ChatGPTReverse a change made through these tools. Pass back the exact undo descriptor the write returned (id, kind, label, payload). Only changes made by this user can be reversed.