assess_data_quality
ChatGPTRun industrial data-quality trust scoring: sensor health, coverage, frequency, outlier, and cross-validation checks. Saves a new analysis/audit record in your Grayson account and counts toward daily analysis usage. It does not delete saved files or replace the loaded dataset.
clear_dataset
ChatGPTRemove the dataset from this user's Grayson workspace. Uploaded data persists across conversations until cleared or replaced — call this when the user wants to start fresh or asks why old data is still loaded.
compute_correlation
ChatGPTCompute a pairwise Pearson or Spearman correlation matrix for specified numeric columns. Returns a chart card for this run. Display each completed result before the overall summary. Saves a new analysis/audit record in your Grayson account and counts toward daily analysis usage. It does not delete saved files or replace the loaded dataset.
compute_cross_correlation
ChatGPTCompute lead/lag cross-correlation between two time-series columns to find temporal relationships. Returns a chart card for this run. Display each completed result before the overall summary. Saves a new analysis/audit record in your Grayson account and counts toward daily analysis usage. It does not delete saved files or replace the loaded dataset.
compute_rolling_correlation
ChatGPTTrack how the correlation between two time-series columns changes over time using a rolling window. Use when the user asks whether a relationship is stable, drifting, or broke down at some point. Returns a chart card for this run. Display each completed result before the overall summary. Saves a new analysis/audit record in your Grayson account and counts toward daily analysis usage. It does not delete saved files or replace the loaded dataset.
detect_anomalies
ChatGPTDetect anomalies in a time-series column using a single method (Isolation Forest, Z-Score, or IQR). Returns a chart card for this run. Display each completed result before the overall summary. For anomaly detection on multiple sensors or a whole file, use detect_anomalies_batch instead so all charts stay together. Saves a new analysis/audit record in your Grayson account and counts toward daily analysis usage. It does not delete saved files or replace the loaded dataset.
detect_anomalies_all_methods
ChatGPTRun Isolation Forest, Z-Score, and IQR on one column with fixed sensitivity/window. Use when the user asks for all methods or a full anomaly comparison. Returns a chart card for this run. Display each completed result before the overall summary. For anomaly detection on multiple sensors or a whole file, use detect_anomalies_batch instead so all charts stay together. Saves a new analysis/audit record in your Grayson account and counts toward daily analysis usage. It does not delete saved files or replace the loaded dataset.
detect_anomalies_batch
ChatGPTRun anomaly detection across the dataset's sensors in ONE call. Use for 'run anomaly detection on this file', all-sensor requests, or multiple columns. Returns every sensor chart together in one results widget, so intermediate charts cannot be replaced. Defaults to Isolation Forest on all non-binary numeric columns; excluded event indicators are reported. Use target_columns to select specific columns, including binary indicators if requested. Returns a chart card for this run. Display each completed result before the overall summary. Saves a new analysis/audit record in your Grayson account and counts toward daily analysis usage. It does not delete saved files or replace the loaded dataset.
export_audit_trail
ChatGPTExport the audit trail for this session: every analysis recorded so far, with the dataset SHA-256 fingerprint, the deterministic functions invoked, all parameters including defaults, the time window, and a SHA-256 hash of the results. Use when the user asks to verify, audit, or prove an analysis. Returns the existing audit trail inline without changing the workspace or stored analyses; downloadable exports are disabled.
generate_sample_data
ChatGPTGenerate synthetic industrial time-series demo data with optional anomalies so users can try analysis features without uploading a file. All parameters are optional with sensible defaults (365 daily rows, generic industry, anomalies included, 4 columns) — call with no arguments unless the user asked for something specific; never ask which options they want. Profiles: generic industrial, drilling (WOB, RPM, ROP, torque, ...), steel_welding (PEB built-up member SAW line: weld current/voltage, heat input, ...) and steel_coating (shot-blast/paint line: DFT, booth temp, viscosity, ...). Supports 4, 8, 12, 16 or 20 data columns. Changes the current workspace and can replace the loaded dataset; existing saved files and analyses are retained.
get_account
ChatGPTShow the user's Grayson plan and usage: the plan, today's usage against its daily limit (analyses run here and Teams messages share one budget) and when it resets, analyses in the last 30 days, files stored, how long files and records are kept, and member since.
get_analysis
ChatGPTShow one of the user's saved analyses again: its result card with the recorded charts, plus parameters, window, summary and record link. Nothing is recomputed — results are exactly as recorded. Pass analysis_id (from list_analyses or a record link) or 'latest' for the most recent; with 'latest', kind picks the newest of that kind (e.g. anomalies). A question that quotes a record id is about that analysis — for 'the point at <time> on <sensor> in record <id>', pass the id, the time as at and the sensor as column.
get_statistics
ChatGPTReturn descriptive statistics (mean, median, std, min, max, skewness) for one or more columns. Saves a new analysis/audit record in your Grayson account and counts toward daily analysis usage. It does not delete saved files or replace the loaded dataset.
get_workspace
ChatGPTReport what is loaded in the user's Grayson workspace right now: the file, rows, columns, numeric columns, time span and when it was loaded. Call it when the user asks what is loaded, or to learn the column names before an analysis instead of asking the user for them.
list_analyses
ChatGPTList the user's saved analyses, newest first — from ChatGPT, Claude or Teams: what ran, on which file, when, a one-line result and the record link. Narrow with kind, file_id (from list_files) or since. get_analysis shows one again with its charts.
list_files
ChatGPTList the files this user has loaded into Grayson — from ChatGPT, Claude or Teams — newest first: name, rows and columns, when it was uploaded, until when it is kept, how many analyses ran on it, and which one is loaded now. Use it when the user asks about their files or wants earlier data again; open_file then loads one without a re-upload.
open_file
ChatGPTLoad one of the user's saved files back into the workspace, replacing what is loaded, so it can be analysed again without re-uploading. Pass file_id from list_files, or name to open the newest file whose name contains it. Shows the dataset card. If the user's message already names an analysis, pass it as analysis_requested and run that analysis right after. Changes the current workspace and can replace the loaded dataset; existing saved files and analyses are retained.
run_forecast
ChatGPTForecast future values of a time-series column. Simple Moving Average smooths recent history, Linear Trend fits a straight-line trend, Prophet models trend plus seasonality. Returns a chart card for this run. Display each completed result before the overall summary. Saves a new analysis/audit record in your Grayson account and counts toward daily analysis usage. It does not delete saved files or replace the loaded dataset.
show_welcome
ChatGPTShow Grayson's capability overview card. Call this when the user greets Grayson or asks what Grayson can do — before answering in your own words. For an analysis request, call the analysis tool instead; this card does not show results. Also use it for the direct Upload file fallback described by upload_dataset_file. Takes no arguments.
upload_dataset
ChatGPTLoad pasted CSV text into this session. For attached CSV/XLSX files use upload_dataset_file with the original file reference; never convert spreadsheets to CSV. For pasted text, call this immediately — never ask what to analyse first: the card this returns offers the analyses, unless the message already named one (see analysis_requested). Pass the file's text content exactly as-is: the time column is detected and parsed automatically whatever it is named (timestamp, time, datetime, ...) — never rename headers or otherwise modify the user's data. NEVER sample, truncate, or thin the user's data to fit a size limit: Grayson's results are audited against the exact dataset, and silently reduced data invalidates them. For files too large for one call, send the WHOLE file across several calls: the first call as normal, every later call with append=true, each chunk starting with the same header row. If the file is too large to relay as text at all, call show_welcome and tell the user to press its Upload file button — the card uploads the file directly, bypassing this limit entirely. The response reports total rows loaded so far — verify it matches the file before analysing. Changes the current workspace and can replace the loaded dataset; existing saved files and analyses are retained.
upload_dataset_file
ChatGPTLoad an attached CSV or XLSX file directly from its original file reference. Call this immediately for spreadsheet attachments. Pass the host-provided file object; DO NOT read, convert to CSV, base64-encode, sample, or relay the workbook through model text. Grayson downloads and parses the original bytes. If the host cannot supply a file reference, show the welcome card and ask the user to use Upload file. Defaults to the first workbook sheet and reports all sheet names; set sheet_name to select another. content_base64 is only for bytes sent by the widget's file picker. Changes the current workspace and can replace the loaded dataset; existing saved files and analyses are retained. A file reference fetches bytes from its supplied public HTTPS host and can save the file in your Grayson library.