collect_feedback
ChatGPTCollects user feedback on the provided response. When to use this tool: - After providing an analysis, a SQL query, or an important response - When you want to know if the response was helpful - Naturally suggest: "Was this response helpful? 👍 👎" Ratings: - 'positive': The response was helpful and accurate - 'negative': The response was not satisfactory - 'neutral': Neither satisfied nor dissatisfied Categories (optional): - 'accuracy': Was the response accurate? - 'relevance': Did the response address the question? - 'completeness': Was the response complete? - 'speed': Was the response time acceptable? - 'other': Other feedback Feedback usage: Feedback is used to improve future responses (RAG, analytics).
configure_agent
ChatGPTAjuste un agent déjà activé : ses instructions (context) et/ou ses actions (couper une alerte, changer un seuil). Chaque action a un id et des params (lus dans list_agents) ; ne transmettre QUE les actions à modifier, avec seulement les params à changer. Toute modification fait réécrire le plan au prochain passage (quelques dizaines de minutes). Écriture : indisponible en mode clé API.
create_aggregation_view
ChatGPTCreates a materialized view or stored procedure in the project's BigQuery data warehouse for data pre-aggregation. When to use this tool: - When the user needs to pre-aggregate data from multiple connectors (e.g., cross-channel marketing report) - When a query is too slow to run on-demand and benefits from materialization - When the user asks to "create a view", "save this as a table", "materialize this query" Naming rules (enforced): - Target dataset MUST be 'quanti_agg' (created automatically if it doesn't exist) - Object name MUST start with 'llm_' prefix (e.g., llm_weekly_spend) - Format: CREATE MATERIALIZED VIEW quanti_agg.llm_name AS SELECT ... SQL format: - CREATE MATERIALIZED VIEW: for pre-computed aggregation tables - CREATE OR REPLACE MATERIALIZED VIEW: to update an existing view - CREATE PROCEDURE: for complex multi-step transformations Example: CREATE MATERIALIZED VIEW quanti_agg.llm_weekly_channel_spend AS SELECT DATE_TRUNC(date, WEEK) as week, channel, SUM(spend) as total_spend FROM prod_google_ads_v2.campaign_stats GROUP BY 1, 2 Limits: Maximum 20 active aggregation views per project.
create_dashboard
ChatGPTCrée et persiste un dashboard à partir d'une spec déclarative Quanti (pages, blocs : heading/text/callout/section/grid/kpi_band/kpi_card/chart/table) et de ses bindings de données (queries clé→SQL). Référence un template via template_slug. N'émets JAMAIS de HTML ni de Chart.js : uniquement la spec JSON. ⚠️ Appelle get_dashboard_guide AVANT : le SQL de chaque binding doit renvoyer une forme tabulaire précise selon le bloc (kpi=1 ligne, chart=col0 labels + colonnes séries, table=1 colonne par column.key).
create_template
ChatGPTCrée un template de dashboard (charte réutilisable : couleurs, typo, logo, header/footer) pour un projet. Suit le schéma Quanti template (version='1', tokens.colors.primary/accent, ...). Plusieurs templates par projet ; un dashboard référence un template par son slug. À créer avant le premier dashboard si aucun n'existe. Appelle d'abord get_dashboard_guide pour le format exact.
create_use_case
ChatGPTCreates and saves a new use case (reusable analysis). When to use this tool: - When the user asks to "save this analysis", "create a use case", "remember this query" - After building a SQL query the user wants to reuse - To capitalize on a recurring business analysis Available scopes: - 'member' (default): Personal use case, visible only to you - 'project': Shared with the entire project team (requires project_id) Best practices: - Slug: technical identifier in snake_case (e.g., weekly_campaign_performance) - Name: human-readable name (e.g., "Weekly Campaign Performance") - Description: explain the business context and when to use this analysis - SQL template: include the SQL query if it's generic and reusable
delete_aggregation_view
ChatGPTDeletes an aggregation view (materialized view or procedure) from the project. When to use this tool: - When the user explicitly asks to delete/drop a view - To clean up unused or obsolete aggregations - When the project has reached the maximum number of views (20) Warning: This marks the view as dropped in Quanti's tracking. The actual BigQuery object may need manual cleanup. Tip: Use list_aggregation_views first to get the view ID.
delete_use_case
ChatGPTPermanently deletes a use case you created. When to use this tool: - When the user explicitly asks to delete a use case - To clean up obsolete or duplicate use cases Warning: This action is irreversible. The use case will be permanently deleted. Permissions: You can only delete use cases you created. Tip: Ask for user confirmation before deleting.
enable_agent
ChatGPTActive (ou met en pause) un agent sur le projet. À l'activation, l'agent démarre avec ses actions par défaut et écrit son plan dans la journée ; un context optionnel lui donne ce qui ne se lit pas dans les données (budget média mensuel, objectifs, périmètre à exclure). Jamais refusé faute de données : l'agent tourne avec ce qui existe, et son plan dit ce qui lui manque et quoi brancher (missing_sources, aussi poussé en recommandations). DEMANDER confirmation avant de mettre un agent en pause. Écriture : indisponible en mode clé API.
estimate_dashboard_cost
ChatGPTEstime le coût (octets BigQuery scannés, sans exécuter ni facturer) des requêtes d'un dashboard, AVANT de le créer. Passe les mêmes queries que create_dashboard. À appeler pour informer l'utilisateur du coût récurrent avant validation.
execute_query
ChatGPTExecutes a read-only SQL SELECT query on the project's BigQuery data warehouse. No data modification allowed. Table format: Use dataset.table (e.g., prod_google_ads_v2.campaign_stats). Do NOT prefix with a project_id.
get_agent_plan
ChatGPTLe plan d'un agent : ce qu'il surveille concrètement, KPI par KPI, avec l'action qu'il sert, le pourquoi et le statut (ok / invalide avec la raison). Le SQL de chaque KPI n'est renvoyé que si include_sql=true — ne le demander que si l'utilisateur veut voir les requêtes.
get_check_history
ChatGPTReturns the recorded results of the project's data quality checks: per check, table and date, the value, baseline and status. When to use this tool: - In the daily run, after record_check_results: compare today's rows with the previous days (is this a new problem, or the same one as yesterday?). - The user asks "how did my checks do this month?". Empty history = no check is set up on this project. Do not audit the data by hand instead: propose the data quality journey (get_use_cases with the user's need as the question, category "Activation"). Filters: check_id and table are optional. Range: 120 days maximum, default the last 30 days.
get_connector_setup_status
ChatGPTReturns the state of a connector setup session opened by setup_connector: the steps already done, the next step (next), and what that step needs: the sign-in link (connect_url), the accounts to choose from, or the report templates to choose from. Read-only: changes nothing. Not available with an API key.
get_dashboard
ChatGPTRécupère la spec ET les queries ACTUELLES d'un dashboard. À appeler AVANT update_dashboard : lis la spec, modifie-la, renvoie la version complète.
get_dashboard_guide
ChatGPTGuide de création de dashboards Quanti : vocabulaire des blocs, CONTRAT DONNÉES (quelle forme tabulaire ton SQL doit renvoyer par type de bloc), bonnes pratiques de structure, et exemple complet. À appeler AVANT create_dashboard/create_template.
get_help
ChatGPTSearches the official Quanti documentation (docs.quanti.io) to answer questions about using the platform. When to use this tool: - When the user asks "how to do X in Quanti?", "what is a connector?", "how to configure BigQuery?" - When the user needs help configuring or using a connector (Google Ads, Meta, Piano, etc.) - To explain Quanti concepts: projects, connectors, prebuilds, data warehouse, tag tracker, transformations - When the user asks about the Quanti MCP (setup, overview, semantic layer) This tool does NOT replace: - get_schema_context: to get the actual BigQuery schema for a client project - list_prebuilds: to list pre-configured reports for a connector - get_use_cases: to find reusable analyses - execute_query: to execute SQL Available topic filters: connectors, data-warehouses, data-management, tag-tracker, mcp-server, transformations
get_launch_context
ChatGPTRetrieves the full context of a Quanti launch session. The user has pre-configured an analysis from the Quanti interface and was redirected here with a launch_id. Call this function to get the analysis details to execute (name, prompt or SQL template, project).
get_project_context
ChatGPTGets the context of a project (active connectors, available datasets, branding). Use the folderId obtained from list_projects. The response includes a 'branding' object (logo_url, primary_color, secondary_color, tertiary_color, font_family) when configured. Always call this tool before generating a report to apply the project's visual identity.
get_schema_context
ChatGPTStep 1 of schema discovery: returns the catalog of tables relevant to the user's question. Each table comes with its dataset, business name, dw_table_name and a short description — but NOT the field-level details (no columns, no types, no semantic codes). Use the catalog to identify the most promising candidate(s), then call get_table_schema to fetch the full structure of a specific table before writing SQL. IMPORTANT for SQL queries: Use ONLY the dataset.table format (e.g., prod_google_ads_v2.campaign_stats). NEVER prefix with a project_id.
get_sync_history
ChatGPTReturns the Quanti sync history of a warehouse table: for each date, whether the connector sync ran, how many rows it wrote, and why it failed. When to use this tool: - A data check found a missing day, a drop or a stale table, and you need its cause. - The user asks "did the sync run?", "why is this day empty?", "is my connector working?". How to read the answer (per report and date): - ever_succeeded true and max_rows_written 0: the sync ran fine and the provider had nothing for that day. A quiet day, not a hole — unless the date is recent: providers often deliver a day later, and the sync re-reads recent days. - last_status "failed": the sync failed. cause says why: auth_expired, auth_rejected, account_access_lost, rate_limited, provider_unavailable, account_limitation, warehouse_write_failed, cost_limit, timeout, unknown. - last_status "no_pass": no sync covered this date at all. Look at the account: paused, or its last failure. - date empty: a report without dates (reference data), synced as a whole. Rules: - State a cause only if this tool shows it. unknown means unknown: say "cause unknown, check the connector in Quanti". Never guess. - resolved_by "table_name" means the table is not in the semantic layer and its report was matched by name: say the link is assumed. - Range: 90 days maximum. Default: the last 30 days.
get_table_schema
ChatGPTStep 2 of schema discovery: returns the full structure of a single table — fields with types, semantic codes, business names, and the semantic definitions referenced by those codes. Call this AFTER get_schema_context once you've picked a candidate table. Returns: dw_table_name (use this in SQL), description, fields[], semantic_definitions{} keyed by code.
get_use_cases
ChatGPTSearches for relevant use cases to answer the user's question. Use cases contain SQL templates and business definitions. Use this function to discover available analyses.
inspect_reverse_push
ChatGPTReads the pushes of a reverse connector (data sent from the warehouse to Meta, Google Ads, GA4…). Read-only: nothing is saved or sent. Actions: - templates: the connector's push templates (push_id, destination object, sync mode). - describe (push_id, optional account_id): the destination fields the template expects (required, hashed by Quanti…) and the configuration in place. - preview (account_id, push_id + the configuration to test): what the next run would read from the source and send, with warnings (an empty window first) and personal data masked. Each push declared consumes one Activate unit of the user's plan, except on reverse-sgtm-store (enrichment store of a sGTM server, included in its option: account_id = server id, name = collection name read by GTM). Not available with an API key.
invite_to_dashboard
ChatGPTInvite une personne (par email) à consulter un dashboard. Réservé à l'owner. L'invité reçoit un email d'invitation et voit aussi l'invitation sur sa home Quanti. Utilise list_dashboards pour trouver l'id.
list_agent_recommendations
ChatGPTLes recommandations des agents du projet : ouvertes par défaut (new, acked, snoozed), ou toutes. Chaque recommandation a un identifiant (les 8 premiers caractères suffisent pour review_agent_recommendation), un agent, une sévérité, un statut, un titre et un corps rédigé pour le client. Les synthèses des managers (CMO, CEO) ont kind=synthesis.
list_agents
ChatGPTListe l'organisation d'agents IA du projet : l'organigramme (CEO → directions → spécialistes), la mission de chacun en une phrase, les données qu'il attend et celles que le projet a déjà (requirements, cochés), si des données essentielles manquent (applicable=false, avec la raison), s'il est actif ou en pause, et l'état de son plan. TOUJOURS appeler ce tool en premier avant d'activer, configurer ou interroger un agent : il donne les slugs. Les agents tournent seuls (plan écrit par le modèle sur les données du client, puis exécuté chaque jour) et poussent des recommandations.
list_aggregation_views
ChatGPTLists aggregation views (materialized views and procedures) created for a project. When to use this tool: - When the user asks "what views exist?", "my aggregations", "my materialized views" - Before creating a new view to check it doesn't already exist - To get the view ID for deletion Response format: Returns a JSON array with each view's ID, full_name (dataset.name), type, SQL, description, and creation date.
list_connector_reports
ChatGPTLists the catalogue of standardized reports Quanti maintains for a given connector (also called "prebuilds" inside Quanti). What is one of these reports? It is a standardized report maintained by Quanti for a given connector (e.g., Campaign Stats for Google Ads). It defines the BigQuery table structure (columns, types, metrics) and the associated API query. For a REVERSE connector, there is no report and no table: the templates are outbound push contracts (which fields the third-party API expects). The response shape differs accordingly — see Response format below. When to use this tool: - When the user asks "what reports are available for [connector]?" - When the user doesn't know which data or metrics exist for a connector - BEFORE get_schema_context, to explore available reports for a connector - To understand the data structure before writing SQL Difference with get_schema_context: - list_connector_reports → discover which reports/tables EXIST for a connector (catalog) - get_schema_context → get the actual BigQuery schema for the client project (effective data) Response format — TWO possible shapes, check which one you got: 1. Inbound connectors (the usual case) → key "prebuilds". For each: ID, name, description, BigQuery table name, and fields (name, type, description, is_metric). Fields with is_metric=true are aggregatable metrics (impressions, clicks, cost...), the others are dimensions (date, campaign_name...). These tables exist in the warehouse: you can write SQL against them. 2. Reverse connectors (SKU starting with "reverse-", e.g. reverse-googleads, reverse-meta) → key "pushes" and kind="reverse_pushes". These connectors SEND data OUT to a third-party platform (Google Ads, Meta...). Each push lists the destination fields to be FED, with: field_id, type, required, is_primary_key, and the transforms Quanti applies (auto_hash for PII hashed in SHA256, auto_normalize_phone for E.164). ⚠️ These field_id are NOT warehouse columns — no table exists behind them. Never write SQL against them. Use them to answer "what do I need to provide to feed this destination?", to help build the source table, or to explain a mapping. The mapping itself (destination field → source column) is configured in the connector's Mapping tab. SKU examples: googleads, meta, tiktok, tiktok-organic, amazon-ads, amazon-dsp, piano, shopify-v2, microsoftads, prestashop-api, mailchimp, kwanko — and for reverse: reverse-googleads, reverse-meta, reverse-meta-pixel, reverse-tiktok, reverse-amazonads, reverse-adobe
list_dashboards
ChatGPTListe les dashboards d'un projet auxquels l'utilisateur a accès (id, titre, url). À utiliser pour retrouver un dashboard à modifier.
list_my_use_cases
ChatGPTLists your personal use cases (scope: member). What is a use case? A use case is a reusable analysis you created or saved. It contains a business description and optionally a SQL template. When to use this tool: - When the user asks for "my analyses", "my use cases", "what I saved" - Before creating a new use case to check it doesn't already exist - To find the ID of a use case to modify or delete Visibility: These use cases are private and only visible to you.
list_prebuilds
ChatGPTLists pre-configured reports (prebuilds) available for a connector. What is a prebuild? A prebuild is a standardized report maintained by Quanti for a given connector (e.g., Campaign Stats for Google Ads). It defines the BigQuery table structure (columns, types, metrics) and the associated API query. When to use this tool: - When the user asks "what reports are available for [connector]?" - When the user doesn't know which data or metrics exist for a connector - BEFORE get_schema_context, to explore available reports for a connector - To understand the data structure before writing SQL Difference with get_schema_context: - list_prebuilds → discover which reports/tables EXIST for a connector (catalog) - get_schema_context → get the actual BigQuery schema for the client project (effective data) Response format: Returns a JSON with for each prebuild: its ID, name, description, BigQuery table name, and the list of fields (name, type, description, is_metric). Fields marked is_metric=true are aggregatable metrics (impressions, clicks, cost...), others are dimensions (date, campaign_name...). SKU examples: googleads, meta, tiktok, tiktok-organic, amazon-ads, amazon-dsp, piano, shopify-v2, microsoftads, prestashop-api, mailchimp, kwanko
list_project_use_cases
ChatGPTLists use cases shared with the project team (scope: project). When to use this tool: - When the user asks for "team analyses", "project use cases" - To see what colleagues have shared - Before sharing a new use case to avoid duplicates Visibility: These use cases are visible to all project members.
list_projects
ChatGPTLists all projects accessible by the user. Call this function first to discover available projects.
list_scheduled_queries
ChatGPTLists scheduled queries configured in the project's BigQuery. What is a scheduled query? A scheduled query is a SQL query automatically executed on a defined schedule in BigQuery. It is used to aggregate data, populate reporting tables, or perform recurring transformations. When to use this tool: - When the user EXPLICITLY asks about scheduled queries, BigQuery pipelines, or transfer configs - When the user asks "what are my scheduled queries?", "my BigQuery pipelines" - To check execution frequency or status of a specific scheduled query Do NOT use this tool for general data queries, analytics, or when the user asks about page views, metrics, or data from connectors. Use get_schema_context + execute_query instead. Available filters: - dataset: filter by destination dataset (e.g., 'prod_reports') - status: filter by status 'active' (enabled) or 'disabled' Response format: Returns a JSON with for each scheduled query: its name, SQL query, execution schedule, destination dataset, status (active/disabled), and last/next execution dates.
list_templates
ChatGPTListe les templates (chartes graphiques) EXISTANTS du projet (slug, nom, is_default + spec couleurs/typo). ⚠️ APPELLE-LE AVANT toute création OU modification graphique : réutilise un template existant (référence son slug dans create_dashboard) ou modifie-le (update_template) — ne crée un nouveau template (create_template) QUE si la liste est vide.
record_check_results
ChatGPTRecords the daily results of ONE data quality check on ONE table, so later runs can compare against them. When to use this tool: - After the backtest of a check the user kept: record its rows, they become the history. - In the daily run, right after running a saved check's SQL. Rules: - check_id is the ID of the check saved as a project use case (save_use_case, scope='project'). Save the check first. - One row per date, exactly as the check's SQL returns it: date, value, baseline, status. Numbers only: no text, no client data. - status is one of: ok, pending, not_applicable, missing, low, high, inconsistent. - A date already recorded for this check and table is replaced (a "pending" day can become "ok"). - 200 rows per call at most; dates within the last 120 days. Older results are deleted.
review_agent_recommendation
ChatGPTTraite une recommandation : ack (vu, je m'en occupe), snooze (me la remontrer dans N jours), dismiss (plus jamais celle-ci). Un verdict et un commentaire sont relus par l'agent à sa prochaine planification : c'est ainsi que le client corrige un seuil ou un faux positif. DEMANDER confirmation avant dismiss, et ne jamais inventer d'identifiant : le prendre dans list_agent_recommendations ou dans le mail reçu. Écriture : indisponible en mode clé API.
run_marketing_mix_model
ChatGPTRe-runs a Marketing Mix Modeling study previously configured with setup_marketing_mix_model. Important: Do NOT call this right after setup_marketing_mix_model. The first run is automatically triggered by setup_marketing_mix_model. Use run_marketing_mix_model only to re-launch an existing study later (e.g., after data refresh or parameter changes). Prerequisite: Must have called setup_marketing_mix_model first to obtain an account_id. Duration: The Meridian fit (MCMC) takes approximately 10-30 minutes depending on data volume. The user will receive an email when results are ready. Results: Results are written to the project's data warehouse (mmm_channel_summary and mmm_weekly_contributions tables). They can then be queried via execute_query.
run_mmm
ChatGPTRe-runs a Marketing Mix Modeling study previously configured with setup_mmm. Important: Do NOT call this right after setup_mmm. The first run is automatically triggered by setup_mmm. Use run_mmm only to re-launch an existing study later (e.g., after data refresh or parameter changes). Prerequisite: Must have called setup_mmm first to obtain an account_id. Duration: The Meridian fit (MCMC) takes approximately 10-30 minutes depending on data volume. The user will receive an email when results are ready. Results: Results are written to the project's data warehouse (mmm_channel_summary and mmm_weekly_contributions tables). They can then be queried via execute_query.
save_dashboard
ChatGPTCrée un dashboard, le modifie, ou estime son coût, à partir d'une spec déclarative Quanti (pages, blocs : heading/text/callout/section/grid/kpi_band/kpi_card/chart/table) et de ses bindings de données (queries clé→SQL). N'émets JAMAIS de HTML ni de Chart.js : uniquement la spec JSON. - Estimer (dry_run=true, queries requis) : octets BigQuery scannés et coût, sans rien exécuter ni créer. À faire AVANT la création pour informer l'utilisateur du coût récurrent. - Créer (sans dashboard_id) : title, slug et spec requis. - Modifier (avec dashboard_id) : garde l'id, l'URL et le partage. Remplace la spec (et title/template/queries si fournis) : get_dashboard d'abord, renvoie la spec COMPLÈTE. Réservé à l'owner/editor.
save_template
ChatGPTCrée ou modifie un template de dashboard (charte réutilisable : couleurs, typo, logo, header/footer), identifié par son slug dans le projet. ⚠️ Si un template porte déjà ce slug, il est MODIFIÉ et tous les dashboards qui l'utilisent récupèrent la nouvelle charte : c'est ce qu'il faut pour toute modification graphique (slug du template en cours, lu via get_dashboard). Un nouveau slug crée un nouveau template, réservé à une NOUVELLE palette — propose alors les couleurs à l'utilisateur. Fournis la spec COMPLÈTE (format dans le guide de list_dashboards).
save_use_case
ChatGPTSaves a use case (reusable analysis): creates it, or updates it when use_case_id is given. When to use this tool: - When the user asks to "save this analysis", "create a use case", "remember this query" - After building a SQL query the user wants to reuse - To improve, fix or recategorize an existing use case (pass its use_case_id; only the fields given change) Available scopes (creation only): - 'member' (default): Personal use case, visible only to you - 'project': Shared with the entire project team (requires project_id) Permissions: You can only update use cases you created. Best practices: - Call get_use_cases (list mode) first: to avoid a duplicate, and to get the ID of the one to update - Slug: technical identifier in snake_case (e.g., weekly_campaign_performance) - Name: human-readable name (e.g., "Weekly Campaign Performance") - Description: explain the business context and when to use this analysis - SQL template: include the SQL query if it's generic and reusable
search_documentation
ChatGPTSearches the official Quanti documentation (docs.quanti.io) to answer questions about using the platform. When to use this tool: - When the user asks "how to do X in Quanti?", "what is a connector?", "how to configure BigQuery?" - When the user needs help configuring or using a connector (Google Ads, Meta, Piano, etc.) - To explain Quanti concepts: projects, connectors, prebuilds, data warehouse, tag tracker, transformations - When the user asks about the Quanti MCP (setup, overview, semantic layer) This tool does NOT replace: - get_schema_context: to get the actual BigQuery schema for a client project - list_connector_reports: to list the standardized reports available for a connector - get_use_cases: to find reusable analyses - execute_query: to execute SQL Available topic filters: connectors, data-warehouses, data-management, tag-tracker, mcp-server, transformations
send_feedback
ChatGPTSends the user's feedback about Quanti to the Quanti team. Call this tool ONLY when the user explicitly asks to send feedback, report a problem, or rate a Quanti answer. Do not offer it, do not ask the user for a rating, and never call it on your own initiative after an analysis. What is recorded: the rating, the optional category, the comment the user wrote, and the name of the Quanti tool concerned. Nothing else from the exchange is sent — do not summarise or paraphrase the conversation into the comment, pass the user's own words. Ratings: - 'positive': the user is satisfied - 'negative': the user is not satisfied - 'neutral': neither Categories (optional): 'accuracy', 'relevance', 'completeness', 'speed', 'other'.
setup_connector
ChatGPTCrée un connecteur (pull ou reverse) pas à pas, via une session de setup Quanti. Appeler avec action, dans cet ordre : 1. start (sku) : ouvre la session et renvoie un connect_url. Les credentials (clé d'API, OAuth) ne passent JAMAIS par la conversation : donne ce lien à l'utilisateur, il s'y connecte à Quanti et autorise l'accès, puis revient. Ne lui demande JAMAIS de coller une clé ici. 2. status (session_id) : où en est la session et ce qu'il faut faire ensuite (next) ; renvoie les comptes à choisir quand c'est le moment. 3. select_account (account_id) puis, si demandé, select_sub_accounts (accounts) ; pour un connecteur pull, select_templates (template_ids). 4. create (account_name, sync_hour, lookback_days) : crée le connecteur. Pour un reverse, sync_hour doit être APRÈS l'heure de fin des connecteurs qui alimentent la source (sinon le push lit une table vide, run vert, rien envoyé) ; lookback_days = 2 pour Meta Pixel et GA4 (dédoublonnage 48 h), 7 sinon. Webhook à schéma fixe (ex. webhook-tag-analytics) : start puis create avec account_name seul (ni credentials, ni synchro) ; renvoie webhook_url et uniq_id, à mettre dans le tag. Un webhook qui déduit son schéma d'un payload de test se crée depuis l'app Quanti. Les pushes d'un reverse se configurent ENSUITE avec configure_push. Écriture : indisponible en mode clé API.
setup_marketing_mix_model
ChatGPTConfigures a Marketing Mix Modeling (MMM) study for a project. What is MMM? Marketing Mix Modeling measures the real contribution of each marketing channel (Google Ads, Meta, etc.) on a KPI (leads, revenue, conversions), accounting for external factors (seasonality, holidays, promotions). Recommended workflow: 1. Use get_schema_context to discover the project's tables/columns 2. Generate input SQL queries (KPI, channels, exogenous variables) 3. Validate each query before calling setup_marketing_mix_model: Use execute_query to run a COUNT() wrapper on each input query (e.g., SELECT COUNT() FROM (<query>)). If any query returns 0 rows, do NOT include it in setup_marketing_mix_model — warn the user that the data source is empty and ask whether to proceed without it or fix the query. 4. Call setup_marketing_mix_model with the validated SQL queries — the study is automatically launched after setup 5. Do NOT call run_marketing_mix_model after setup_marketing_mix_model: the first run is triggered automatically Important: run_marketing_mix_model is only needed to RE-RUN an existing study later, not after initial setup. Input queries format: Each query must return a "time" column (DATE) and the requested metrics. - role="kpi": a "kpi" column (the target KPI) - role="channel": "spend" and "impressions" columns + channel_name - role="exogenous": columns named after the exogenous variables + columns[] Granularity: "weekly" is recommended (MMM standard). SQL should aggregate by week. Important: Adapt the SQL dialect to the project's data warehouse type (BigQuery, Snowflake, Redshift).
setup_mmm
ChatGPTConfigures a Marketing Mix Modeling (MMM) study for a project. What is MMM? Marketing Mix Modeling measures the real contribution of each marketing channel (Google Ads, Meta, etc.) on a KPI (leads, revenue, conversions), accounting for external factors (seasonality, holidays, promotions). Recommended workflow: 1. Use get_schema_context to discover the project's tables/columns 2. Generate input SQL queries (KPI, channels, exogenous variables) 3. Validate each query before calling setup_mmm: Use execute_query to run a COUNT() wrapper on each input query (e.g., SELECT COUNT() FROM (<query>)). If any query returns 0 rows, do NOT include it in setup_mmm — warn the user that the data source is empty and ask whether to proceed without it or fix the query. 4. Call setup_mmm with the validated SQL queries — the study is automatically launched after setup 5. Do NOT call run_mmm after setup_mmm: the first run is triggered automatically Important: run_mmm is only needed to RE-RUN an existing study later, not after initial setup. Input queries format: Each query must return a "time" column (DATE) and the requested metrics. - role="kpi": a "kpi" column (the target KPI) - role="channel": "spend" and "impressions" columns + channel_name - role="exogenous": columns named after the exogenous variables + columns[] Granularity: "weekly" is recommended (MMM standard). SQL should aggregate by week. Important: Adapt the SQL dialect to the project's data warehouse type (BigQuery, Snowflake, Redshift).
unregister_aggregation_view
ChatGPTUnregisters an aggregation view from Quanti: it stops being listed, stops being used by the semantic layer, and frees one of the project's 20 slots. IMPORTANT — this does NOT drop the BigQuery object. The materialized view or procedure keeps existing in the data warehouse, keeps being refreshed, and keeps incurring BigQuery costs until someone drops it manually. Tell the user this plainly whenever you call this tool; do not report the view as deleted from their warehouse. When to use this tool: - When the user explicitly asks to remove a view from Quanti - To clean up unused or obsolete aggregations - When the project has reached the maximum number of views (20) Tip: Use list_aggregation_views first to get the view ID.
update_dashboard
ChatGPTModifie un dashboard EXISTANT (garde son id/url/partage) au lieu d'en créer un nouveau. Remplace la spec (et title/template/queries si fournis). Réservé à l'owner/editor. Workflow : get_dashboard → modifier la spec → update_dashboard avec la spec complète + les queries. N'émets JAMAIS de HTML.
update_template
ChatGPTMet à jour la charte (couleurs, typo, logo, header/footer) d'un template EXISTANT, identifié par son slug. À utiliser pour TOUTE modification graphique d'un dashboard : modifie le template en cours (le slug référencé par le dashboard, lu via get_dashboard → spec.template), ne crée PAS un nouveau template. Tous les dashboards qui utilisent ce template récupèrent la nouvelle charte. Fournis la spec COMPLÈTE du template (même format que create_template).
update_use_case
ChatGPTUpdates an existing use case that you created. When to use this tool: - To improve the description or SQL of an existing use case - To fix an error in a use case - To change a use case's category Permissions: You can only modify use cases you created. Tip: Use list_my_use_cases or list_project_use_cases first to get the use case ID.