HelenAI

Calculates request metrics directly in the analytics database and returns only aggregate values, never individual request records. You MUST use this tool for questions asking how many, total, count, sum, average, minimum, maximum, most, least, ranking, or grouped metrics. This includes overall request totals for a client, hospital, location, or date range. Examples: how many requests were made at TGH in total, total steps, average completion time, request count grouped by command, or total steps grouped by command. For grouped results, set groupBy to a supported metadata field such as title, commandcategorytitle, hospital, building, floor, bed, sourcetitle, receivername, or sendername. Use query_requests only when the user needs request-level details or specific request records. For metric questions involving steps or time values, ask about the relevant outliers before calling this tool and add the corresponding outlier filter when needed.

Creates or inserts a hospital request. Treat create, insert, submit, and send as equivalent when the user asks for a new request. Requires the exact command row GUID (Crg) and the selected bed's bed row GUID (Rbrg). Resolve the user's readable bed name through get_locations before calling this tool. This performs an external write; use only when the user has explicitly asked to create, insert, submit, or send the request.

Returns command categories - the grouping a command type belongs to. Columns: Title, DisplayOrder. Use for questions like what command categories exist or how many there are. Use query_requests instead for questions about a specific request's command category.

Returns physical locations - hospitals, buildings, floors, beds, and rooms. Columns: Hospital, Building, Floor, Bed, Rbrg, RoomName, RoomNumber, RoomTypeName. Bed is the readable bed name and Rbrg is the bed row GUID.

Returns request sources - how a request was submitted (e.g. app, phone, tablet, kiosk). Columns: Title. Use for questions like what sources exist or how many sources there are. Use query_requests instead for questions about a specific request's source.

Returns staff/users - the catalog of people who can send or receive requests. Columns: Name, UserType. Use for questions like how many users exist or what types of staff there are. Use query_requests instead for questions about a specific request's sender/receiver.

Returns the clients (also called tenants) the current user can access, each with its real display name and known aliases (short forms, abbreviations). Call this before query_requests or a catalog tool (query_commands, get_users, get_locations, get_sources, get_command_categories) when the user has not specified which client to query or when a client name or alias needs to be resolved. When the user names a client, match it against the Name or any Alias listed here, then pass the exact Name value to the downstream tool. When clarification is needed, ask the user to choose exactly one client for the current question; do not offer all clients as an option.

Returns available command definitions, including command names, command row GUIDs (Crg), categories, and automation or linen attributes. Supports analytics operations that filter or join requests by command. When a user describes a patient, room, or service need without knowing the exact command name, call this tool without filters to return the full command catalog, then have the assistant semantically evaluate the returned command names. The database does not perform semantic matching of the user's wording.

Searches actual command requests and returns individual request records. Use this only when the user needs request-level details, specific request records, or information about the records themselves. NEVER use this tool for questions asking how many, total, count, sum, average, minimum, maximum, most, least, ranking, or grouped metrics. For all such metric questions, use calculate_request_metrics instead; do not load request records just to calculate a metric. The request is the central record; command, location, source, date, and sender/receiver are metadata attached to it. Filter on any of that metadata. For date ranges, use Created when the user asks when requests were created, and use Completed when the user asks when requests were completed. completionScope only controls whether Completed is null or non-null; it does not specify when completion occurred. For compound questions with multiple results, make a separate call for each result and do not reuse a date filter from another part of the question; the same dates may require different date columns. For metric questions, ask about outliers only for the metric requested: Steps -> IsStepsOutlier, TimeToAcceptSeconds -> TimeToAcceptOutlier, and TimeToCompleteSeconds -> TimeToCompleteOutlier. For linen requests, use the applicable linen outlier flag: TimeToAcceptOutlierLinen or TimeToCompleteOutlierLinen. Ask whether to include or exclude the relevant outliers before calling this tool; do not ask about unrelated outliers or guess. To exclude them, add the corresponding filter with operator = and value false. After answering, state whether the relevant outliers were included or excluded. For questions not about a specific request, use the matching catalog tool instead: query_commands, get_users, get_locations, get_sources, or get_command_categories. Resolve relative dates to actual calendar dates before calling.

Sends an email through the configured hospital email service to the signed-in user's email address, optionally with file attachments. Use this only when the user explicitly asks to send or email something. This performs an external side effect. The recipient is taken from the authenticated user's email, so do not ask the user for a recipient address. Provide the subject and message body. Attachments must be supplied as Base64 content with a filename and optional MIME type.