This Product (LLM AI Charts) processes data you configure it to share with third-party large language models (LLMs). By default, requests are routed to one or more third-party LLM services (which may in turn dispatch to upstream model providers); the specific services and routing are a backend implementation detail and may change at any time. You may additionally configure a "Bring-Your-Own" (BYO) connection — your own API key against your own provider account (e.g., your own Anthropic, OpenAI, or Azure OpenAI account) — in which case requests flow directly to your chosen provider under your contract with them. The Product is not intended for personally identifiable information (PII), classified information, or regulated data (e.g., HIPAA, PCI, GDPR special categories); avoid or obfuscate such fields before they enter your Power BI dataset. Any data the Add-in renders is the worksheet range you selected in Excel.

The Add-in has no "Privacy Level" setting, and no setting that can widen what it sends. Every request it makes about YOUR data — a generation and the free "What fits?" list alike — transmits the MEASURED SHAPE of the range you selected and nothing more:

• Dataset shape: column names, data types, and aggregate statistics (e.g., distinct count, blank count, min/max, mean, standard deviation, modal share).
• Your rows are never transmitted, to us or to any LLM provider. The obfuscated data sample that the Power BI visual may send at its highest privacy level has no equivalent here and cannot be switched on. Text columns are described by a FORMAT CLASS rather than by their contents — "title-case words", "email-like", "fixed-length code" — so the model can recognise what a column is without being shown what is in it.
• Two bounded exceptions, so that a chart can order and label itself correctly. Where a column's distinct values match a recognised ordered scale (for example "Strongly disagree" through "Strongly agree"), that list of labels is transmitted, so the axis is drawn in its own order instead of alphabetically. Where a NON-TEXT column is used as a category — a true/false, a year, a date — up to five of its most common values are transmitted, so the chart can say what it is plotting. Both are limited to short category labels; neither carries a measure, a row, or the contents of a text column. The Add-in shows you the exact payload before you send it: open "See what's sent" beside the range it has measured.

An example chart is different: it is pre-built code drawn over a sample table we supply, and the request for one carries the size of the pane and the shape of that sample table, nothing measured from your own data. When using a BYO provider connection, the same content flows to your chosen provider under your contract with that provider, not ours; data handling, retention, and abuse-review policies in BYO mode are governed by your provider's terms.

Your data is rendered locally. The LLM produces only the chart code; that code then runs inside the Add-in, on your machine, against your full worksheet data — which is not transmitted to us or to the LLM provider. The generated code and the settings that produced it are saved in the workbook, so that reopening it redraws the chart with no request to us and no charge. A workbook you send to someone else carries that saved code with it, and the account-link token if one is present, so treat a shared workbook as you would any file holding a credential.

In hosted (non-BYO) mode, the upstream provider's policies apply. For Microsoft Azure AI Foundry models:

• Prompts and outputs are not used for training by Microsoft or accessible to other customers.
• Prompts may be temporarily stored and reviewed by Microsoft only if flagged for abuse or misuse.

Where we route directly to a provider's native API (e.g., the Anthropic Messages API), that provider's published privacy and abuse-review terms apply in place of Microsoft's.

When LLM AI Charts communicates with our web services, information such as the account-link token, model selection, request shape, and feature usage passes through to our infrastructure. We log activity (request metadata, the prompt we construct and the model's response, the LLM-generated code, runtime outcomes, error stack traces, IP address and the per-installation identifier described below) to diagnose issues, reconcile billing, and improve our services. This logging is managed by bizintelligencechampions.com, is retained on servers in the United States, and is not used for model training or accessible to other customers or partners.

Identifiers. The Add-in mints a random identifier once per installation, keeps it in the browser storage Excel provides, and sends it with every request. It is not a credential and carries no name. We use it to tell installations apart in usage reporting, to vary chart choices between one installation's generations, and as one half of the key our unpaid rate limits are counted against.

The Add-in has one access mode: an organizational ("site") license, purchased through your Biz Intelligence Champions account, which collects some personal data. Use of that Account is governed by our general privacy policy, which you can find here. There is no freemium or no-account trial mode in Excel, and no freemium key is created or sent. What the Add-in offers without an Account is the example-chart lane described above, which sends nothing measured from your data.

The Add-in stores no license key and has no place to type one: a workbook holds only the account-link token described below, and an older workbook that still carries encrypted license credentials has them cleared the next time it is linked.

When BYO mode is configured, the BYO API key (Anthropic, OpenAI, or Azure OpenAI) is stored in your Biz Intelligence Champions account record on our servers, encrypted at rest. We use it solely to authenticate requests to your chosen provider on your behalf; you remain responsible for the provider account itself, including key rotation, usage controls, and provider-side billing.

Linking an account opens a sign-in page hosted on bizintelligencechampions.com in a dialog window. Your password is entered on our own page, never in the Add-in, and the Add-in never receives it. What it receives is a single-use, short-lived code that it exchanges for an account-link token; that token is what the workbook stores and what later requests carry. There is no other way in: the Add-in has no field for a license key.

Diagnostics. The add-in keeps a short log of its own activity and sends it to our services so that we can support the product and diagnose failures. That log records events and measurements - the address of the range you selected, row and column counts, the size of the pane, timings, which chart was drawn, the build in use, and any error text - together with the same column names and data types described above. It does not record the contents of your cells.