Artificial Intelligence Usage Notice

1. Purpose and Scope

Wappi Holding LLC ("Wappi") integrates third-party artificial intelligence (AI) technologies to enhance its conversational commerce services. This AI Usage Notice is intended to transparently inform about: (a) which AI technologies are used and for what purpose; (b) what personal data is processed by AI models; (c) the inherent limitations of AI systems; (d) what controls the User has over AI operation; and (e) how associated risks are managed. This notice supplements our Privacy Policy and Data Processing Agreement (DPA).

2. AI Services and Features

Wappi uses artificial intelligence systems for the following features:

• Conversational agents (chatbots): automated systems that answer questions, recommend products via semantic search (RAG), process orders, manage returns, and provide customer support through WhatsApp, Telegram, and other channels
• Semantic search (RAG — Retrieval-Augmented Generation): finding relevant products based on conversation context, using vector embeddings stored in PostgreSQL with pgvector
• Embedding generation: creating vector representations of products, descriptions, and catalogs to power semantic search
• Intent analysis and classification: automatically determining the purpose of the End Customer's message (inquiry, purchase, complaint, support)
• AI voice agents: making automated calls with synthesized voices for order confirmation, bill collection, appointment reminders, and post-sale follow-up
• Real-time speech transcription (STT): converting spoken audio to text for processing by the conversational agent
• Speech synthesis (TTS): generating spoken audio from the agent's text responses for phone calls
• Content enhancement: automatic product description optimization, tag generation, and intelligent categorization

3. AI Models and Providers

Wappi operates as a technology intermediary connecting Users with AI models developed and operated by third parties. The User can select the provider and model that best suits their needs:

• Free open-source models (no additional AI cost): Llama (Meta), Gemma (Google), Mistral — run via Groq Inc. for accelerated inference
• OpenAI LLC: GPT-4o-mini (default model), GPT-4o (advanced model) — for conversational generation and embeddings (text-embedding-3-small)
• Anthropic PBC: Claude Sonnet — for complex reasoning and advanced conversations
• Google LLC: Gemini Pro/Flash — for voice pipeline processing and conversations
• OpenRouter Inc.: access to additional models from multiple providers under a unified API
• Deepgram Inc.: Nova-3 — for real-time voice transcription (STT) with native Spanish support
• Google Cloud TTS: neural Journey voices — for voice synthesis in phone agents
• ElevenLabs Inc.: high-fidelity voices — premium voice synthesis (optional)
• Cartesia AI: low-latency streaming voices — real-time voice synthesis (optional)

5. Risk Classification and Regulatory Compliance

Regarding applicable artificial intelligence regulation:

• EU AI Act (Regulation EU 2024/1689): Wappi's conversational agents are classified as limited-risk AI systems under Article 50, subject to transparency obligations. Wappi complies with these obligations by informing End Customers that they are interacting with an automated system
• Transparency: We strongly recommend that Users configure an introductory message informing End Customers that they are interacting with an AI assistant
• No decisions with legal effect: Wappi's AI Agents are not designed to make decisions that produce legal effects or that significantly affect End Customers within the meaning of Art. 22 GDPR
• Human oversight: The Platform design ensures that the User maintains the ability to supervise, correct, and deactivate AI Agents at all times

6. AI Limitations and Risks

The User acknowledges and accepts the following inherent limitations of artificial intelligence systems:

• Hallucinations: AI models may generate factually incorrect information presented with apparent confidence, including erroneous prices, inaccurate availability, or unsuitable recommendations
• Not professional advice: AI outputs do not constitute professional advice of any kind (legal, medical, financial, tax, accounting, or otherwise)
• Variability: performance may vary significantly depending on language, context, query complexity, and selected model
• Bias: AI models may contain inherent biases derived from their training data, which may result in inequitable responses in certain contexts
• No warranty: Wappi does not guarantee the accuracy, completeness, adequacy, timeliness, or reliability of AI-generated responses for any specific purpose
• User responsibility: the User is solely responsible for reviewing, configuring, and supervising AI Agent behavior, and for the consequences of automated interactions with their End Customers

7. User Controls Over AI

Wappi provides the User with granular control over AI features:

• Enable or disable AI Agents at any time, per channel or globally
• Select the preferred AI model (including free open-source models)
• Configure the agent's tone, personality, language, and behavior rules
• Define the product catalog and knowledge base available to the agent
• Set escalation rules to human agents when the AI cannot resolve the query
• Review and audit all agent-managed conversations in real time
• Configure default responses for sensitive or high-risk queries
• Request deletion of AI-processed data in accordance with our Privacy Policy

8. Transparency Toward End Customers

Pursuant to the transparency obligations of the EU AI Act (Art. 50) and as a market best practice, Wappi strongly recommends that Users include a clear and unambiguous notification to End Customers that they are interacting with an AI-assisted system. Wappi provides tools to configure automatic disclosure messages at the start of each conversation.

9. Contact

For questions, requests, or concerns about the use of artificial intelligence at Wappi: [email protected]