AI Features Addendum

This Addendum governs your use of AI-powered features within Atellier ("AI Features"), including the Carl assistant, AI mockup generation, copy generation, smart suggestions, and any other features that use machine-learning models. It supplements, and is incorporated into, the Terms of Service.

1a. The Carl Assistant

"Carl" is the in-app AI assistant. To answer questions and take actions on your behalf, Carl uses a large language model (LLM) hosted by one of our AI sub-processors (see Sub-Processors) together with a retrieval layer that can read records from your organization's workspace.

• Scope of access. Carl can read records across the apps you use Atellier for — deals, clients, contacts, proposals, invoices, projects, factory orders, messages and conversations, training docs, file and image metadata, and similar operational data. Carl's access is strictly limited to the active organization of the signed-in user and is filtered by that user's per-app role and permissions. Carl cannot read data from other tenants and cannot bypass Row-Level Security.
• On-demand retrieval, not bulk export. Carl retrieves only the records it needs to answer the current prompt. We do not bulk-export your Customer Data to model providers, and no Customer Data is used to train any foundation model (see §2).
• Tool use and actions. When you ask Carl to do something (create a deal, send a message, generate a proposal, schedule a follow-up, update a record), Carl invokes the same APIs your user account would. Every action is attributed to the requesting user and recorded in the activity log.
• Admin controls. Organization admins can disable Carl org-wide in Settings, and can restrict which apps and data Carl is allowed to read by adjusting role permissions. Disabling Carl stops further LLM calls and retrieval against your data.
• Logging. Prompts, retrieved context summaries, and responses are logged within your tenant so admins can audit Carl's behavior, and may be transiently retained by the underlying AI provider for abuse-prevention per that provider's published terms.

1b. Public Carl (Marketing Site Assistant)

A separate, sandboxed "Public Carl" assistant is available on our unauthenticated marketing site to answer questions about Atellier itself. Public Carl has no access to any tenant's Customer Data, no access to authenticated user accounts, and only reads our public marketing, documentation, and help content. Conversations with Public Carl are not associated with any customer organization.

1. Inputs and Outputs

"Inputs" are the prompts, files, and context you submit to an AI Feature. "Outputs" are the responses, text, images, or other content the AI Feature generates from those Inputs.

• You retain ownership of your Inputs (subject to any third-party rights in them).
• As between you and Atellier, you own the Outputs you generate, subject to applicable AI provider terms and applicable law (Outputs may not be copyrightable in all jurisdictions, and similar Outputs may be generated for other users from similar prompts).
• You are solely responsible for verifying Outputs before relying on them or sharing them with anyone.

2. No Training on Customer Data

Atellier contractually requires its AI sub-processors not to train their foundation models on Customer Data, and Atellier itself will not use Customer Data to train foundation models. We may use aggregated, de-identified telemetry (e.g., latency, error rates, anonymized usage patterns) to improve product quality.

3. AI Providers

We route AI requests to one or more providers depending on the feature, model availability, and cost. The current provider list is published on the Sub-Processors page. Image generation in the Mockup Generator is powered exclusively by Google's Nano Banana 2 (Gemini 3.1 Flash Image) via the Lovable AI Gateway. We may change AI providers on 30 days' notice.

3a. Watermarking & Provenance

Images generated by the Mockup Generator are produced by Google's Nano Banana 2 model, which embeds an invisible SynthID watermark and C2PA Content Credentials into every output. These signals identify the image as AI-generated and are detectable by SynthID-aware tools and C2PA-compatible viewers. You may not strip, obscure, or attempt to defeat these provenance signals. When you publish or distribute generated content, you should preserve the embedded credentials so downstream recipients can verify origin.

4. No Warranties on Outputs

AI OUTPUTS ARE PROVIDED "AS IS." THEY MAY BE INACCURATE, INCOMPLETE, OFFENSIVE, OUT OF DATE, OR INFRINGING. ATELLIER DOES NOT WARRANT THAT OUTPUTS ARE FACTUALLY CORRECT, FREE OF BIAS, FIT FOR YOUR PURPOSE, OR NON-INFRINGING. YOU MUST INDEPENDENTLY VERIFY ANY OUTPUT BEFORE RELYING ON IT.

5. Prohibited AI Uses

You may not use AI Features to:

• Create deepfakes, voice clones, or synthetic media of identifiable individuals without their explicit, informed consent;
• Generate political disinformation, election manipulation content, or non-consensual sexual content;
• Impersonate any person, business, or government agency in a deceptive way;
• Make automated decisions producing legal or similarly significant effects (employment, credit, insurance, housing, government benefits) without meaningful human review;
• Provide medical, legal, financial, or other professional advice represented as authoritative without appropriate human oversight and disclaimers;
• Generate content that violates the Acceptable Use Policy or any applicable law;
• Build, train, or improve a competing AI product or service.

6. Transparency & AI Act

If you publish or distribute AI-generated content created with the Service to recipients in the European Union, you are responsible for complying with EU AI Act Article 50 transparency obligations, including disclosing that the content has been artificially generated or manipulated when required.

7. Sensitive Inputs

Do not submit special-category personal data (Art. 9 GDPR), Protected Health Information, payment-card numbers, government identifiers, biometric data, or any other regulated data to AI Features unless we have a written agreement permitting it. Inputs may be transmitted to U.S.-based AI providers; do not submit data you are not authorized to disclose internationally.

Because Carl can read records you have stored in Atellier (see §1a), the same restriction applies to the data you place into your tenant: if you store regulated data inside records Carl is permitted to read, that data may be sent to an AI provider when Carl is invoked. Use role permissions or disable Carl to prevent this.

8. Indemnification

You will defend and indemnify Atellier from claims arising from your Inputs, your use of Outputs, or your distribution of AI-generated content, including IP, privacy, defamation, and consumer-protection claims.

Contact

Questions: hello@meonneon.com

Worldwide Signage Ltd. d/b/a Atellier
8405 W 54th Place
Arvada, Colorado 80002
United States