Artificial Intelligence Disclosure

01 Purpose and Flouvia’s role

This document explains which CORD features use artificial intelligence, what data they receive and what controls they have. “Flouvia”, “Customer”, “CORD” and other defined terms have the meaning given to them in the CORD Terms and Conditions.

Flouvia designs and operates CORD’s artificial intelligence systems and is responsible for their design as their provider. The models it uses are Anthropic’s, through its commercial API. No CORD feature makes automated decisions with legal effects on a person or assesses anyone’s credit risk.

02 Features that use artificial intelligence

• AI quoting: in the editor, the Gmail add-on, the API and Elements on the Customer’s website, turns text, images or PDFs into quote line items, with the Customer’s catalog as context. The result is a draft that a person reviews before sending it. The price of a catalog product always comes from the catalog.
• Public quote builder: does the same for visitors without an account. Cord does not keep what they send.
• AI collections: drafts reminders to debtors under the Collections Notice.
• Assistants that propose: guided setup and the Cord Workflows draft propose changes that a person approves; they are never applied on their own.
• Help panel: answers questions about using CORD.

03 MCP servers

Cord’s MCP server. The Customer can connect its own AI assistants to CORD with a key. Each key has permissions by tool or by group (read, drafts and actions with external effects) and only sees the tools it is allowed. An action with external effects, such as sending a quote to a customer or recording a payment, is not carried out when requested: it remains a request that a person at the Customer approves or rejects in CORD. Each call is logged with the key that made it.

External MCP servers used by Cord’s AI. The Customer can connect third-party MCP servers to the quoting assistant. A new server has no allowed tools: the Customer turns them on one by one. Only a tool the server declares as read-only (readOnlyHint) is treated as read-only; any other is a write tool. While the assistant processes a file or the text of an email, it receives no external tools. Before running a write tool, Cord shows the person the system, the tool and the data it will send, and runs it only if the person confirms; the confirmation works once, expires after 15 minutes and is recorded in the audit log.

04 Transparency

Emails drafted by the collections agent visibly state that they were drafted with artificial intelligence on the creditor’s behalf and carry the technical header X-Cord-AI-Generated. The Customer cannot remove that line when editing the draft. Outside automatic collections mode, what the AI proposes remains a draft or proposal until a person sends, publishes or approves it.

05 Data

Anthropic receives only what each feature needs when it is invoked: the text, images or PDFs sent, the catalog as context when quoting and, in collections, the masked context described in the Collections Notice, without the debtor’s name, email, phone number or the payment link. Anthropic deletes API inputs and outputs after 30 days, subject to the exceptions in its policy, and does not use them to train its models. Cord logs, per organization, the use of AI features (model, operation, consumption and result), without the content sent; for the Help panel and the public quote builder it keeps only technical consumption logs.

The Customer shall not send AI features sensitive data, credentials or secrets that are not needed.

06 Limitations and human review

AI can omit, duplicate or draft something incorrectly. Its outputs are not legal, tax or financial advice. The Customer shall review amounts, recipients, data and links before using an output and, in automatic collections mode, supervise sends.

Anyone can ask a person at Flouvia to review an AI output that affects them by writing to legal@flouvia.com.

07 Security

A file or an email may contain hidden instructions aimed at the model. That is why CORD’s AI receives no external tools when processing that content, every write to a connected system requires human confirmation and actions with external effects on Cord’s MCP server require approval. Attempting to manipulate a model to bypass these controls violates the Acceptable Use Policy.

08 Final provisions

This document is read together with the Privacy Notice, the Collections Notice, the Acceptable Use Policy and the Terms. Governing law and jurisdiction are those in section 16 of the Terms, and amendments follow its section 17.