audit_image
ChatGPTUse this when the user attached or linked a photo and wants to know if it is real, fake, edited, AI-generated, stolen from the internet, a photo of a screen, or re-used (an insurance claim, a marketplace listing, a delivery proof, an ID). Always prefer this over create_inspection for a photo the user already has. Pass the attached file as image (or a public link as image_url). Detects photos of screens or printouts, AI-generated or AI-edited images, images already published online and photos already submitted to the account; with expected_object or condition_aspects it also checks what the photo shows. Uses one Rial credit. Returns the result when ready, or a pending id: then call get_fraud_report with that id. An audit is not proof of a live capture — only when the user needs new evidence, use create_inspection to send a live-capture link.
check_verification
ChatGPTRead the result of one inspection by id. Use it to answer "is it done?" or "what did Rial find?". Returns the verdict once the analysis has run — before that it says so explicitly.
create_database
ChatGPTCreate a reusable database from an already parsed table. Send column names and rows verbatim; this tool does not parse CSV text.
create_inspection
ChatGPTCreate a one-off inspection link, but first ask in plain language what must appear in the photo -> expected_object, what visible condition should be assessed -> condition_aspects, whether the place matters -> expected_location, and how many photos/files are needed -> count. If the operator did not mention validations, ask before creating: without expected_object or condition_aspects, Rial checks that the evidence is authentic but not WHAT it shows or its condition. Example for one installed window: mode="verification", count=2, expected_object="una ventana de aluminio instalada", condition_aspects=["vidrio sin rajaduras","marco sin óxido","sellado completo"], expected_location="obra de Av. Corrientes 1234, CABA". Use verification for new live evidence and audit only for files that already exist.
create_inspection_from_template
ChatGPTCreate ONE single-use inspection link from a published template (public_url_name), for one specific person — unlike the permanent public URL, this link belongs to that person and expires. Use it when the operator already has a template and wants to send a link to someone in particular. Pass identification only as the non-sensitive order or case reference the template requests. Do not pass government-issued identifiers. This creates an inspection and can consume billing. A supplied webhook_url sends lifecycle events to an external endpoint; the template may also configure result emails.
create_sheet_database
ChatGPTCreate a Rial database backed by one authorized Google Sheets tab. Preview it first and use the operator-approved name.
create_template
ChatGPTCreate a reusable public capture link, but first MUST interview the operator in plain language and collect: (1) what the content must show -> expected_object; (2) what the AI should verify about its condition -> condition_aspects; (3) whether the capture must happen at a specific place -> expected_location; (4) how many pieces of content and which types or angles -> steps; (5) who should receive each completed result -> notify_email (configures future delivery of results to that external email address); and (6) whether an order or case needs a business reference -> identification (never a government-issued identifier); if the operator did not mention validations, MUST ASK before creating: a template without expected_object or condition_aspects only verifies that the content is authentic, not WHAT it shows or its condition; complete window example: public_url_name="ventana-instalada", name="Ventana instalada", mode="capture", steps=[{key:"frente",type:"image",description:"Foto de frente de toda la ventana"},{key:"sellado",type:"image",description:"Detalle del sellado"}], expected_object="una ventana de aluminio instalada", condition_aspects=["vidrio sin rajaduras","marco sin óxido","sellado completo"], expected_location="obra de Av. Corrientes 1234, CABA", notify_email="obra@example.com", identification={required:true,label:"Número de orden"}; use create_inspection instead for a one-off request.
get_credit_balance
ChatGPTRead the current account's available photos, remaining included allowance and existing prepaid balance. Does not purchase credits or configure payments.
get_fraud_report
ChatGPTShow the fraud report of one verification or audit by id, as a card: the Rial / Not Rial result, every check that tripped (screen, AI, found online), similar-image candidates, re-used photos, and object, location and condition checks. Use it after audit_image returned a pending id, or when the user asks what Rial found for a verification.
get_import_job
ChatGPTRead the status and row totals of one CSV import job. Terminal states are completed, failed and aborted.
get_template
ChatGPTRead one reusable capture link by the name it was published under (public_url_name). The API has no single-template route, so this searches the authoritative template list.
get_usage
ChatGPTRead completed-verification consumption by calendar month for the current account.
get_verification_events
ChatGPTRead the oldest-first audit timeline for one inspection, including lifecycle event types, timestamps and payloads.
list_databases
ChatGPTList the operator's saved CSV and Google Sheets databases, including ids, columns, row counts and sync status.
list_sheets_connections
ChatGPTList Google Sheets OAuth connections already authorized for this account. This is read-only and never starts OAuth.
list_templates
ChatGPTList the account's reusable capture links, with their complete public URL, mode, status and version. Call this before updating or deleting one so you act on a URL that really exists and see what it currently does.
list_verifications
ChatGPTSearch the account's inspections with filters. Use it to answer "how many are still pending?", "what came back suspicious this week?", or to find an inspection whose id you do not have.
preview_csv_import
ChatGPTParse and validate a CSV import against the published template selected by public_url_name, without creating anything. Always run this before start_csv_import.
preview_sheet
ChatGPTPreview one Google Sheets tab before creating a Rial database. Returns the columns, sample rows and total row count without persisting anything.
start_csv_import
ChatGPTStart the reviewed CSV batch for the published template selected by public_url_name and return its job id. Creates inspections in the background using existing account entitlements; completed checks can consume allowance or prepaid balance. Inspections inherit the template's result-email recipient and can send results externally when completed. Repeating the same CSV bytes for the same account and template returns the existing job. Only call after preview_csv_import and explicit user confirmation of the batch and its effects.
sync_database
ChatGPTRefresh one Google Sheets-backed database from its source. Use the exact id from list_databases; CSV databases cannot sync.
update_database
ChatGPTRename a saved database or replace its complete CSV table while preserving its id. A table replacement requires name, columns and rows together.
update_template
ChatGPTComplete or change an existing capture link, including missing object, condition, location, result-email, business-reference (identification), or table settings. Result-email changes configure future delivery to an external address. Use identification only for non-sensitive order or case references; do not request government-issued identifiers. Use get_template or list_templates first, point out every validation warning they return, and offer to fill the gap. Send only fields the operator wants changed: touching steps or mode publishes a new version, and a status change is always sent alone.