We build AI agents that talk to real customers, so we take their behavior seriously. These are the principles we apply on every project.
On this page
A person handles what's sensitive
On every project, someone on the client's team reviews sensitive cases and decides when a conversation is handed over to a human. The agent helps; it doesn't replace your team's judgment.
The agent says what it is
Our agents identify themselves as assistants. They don't pretend to be people.
Our own WhatsApp
When you message Taveras Solutions on WhatsApp, an AI agent replies first. If needed, it hands the conversation to Erik or Raysa. If you'd rather talk to a person from the start, ask for one in the same chat or email us at erik@taverassolutions.com.
Your data doesn't train models
We don't train models on a client's data without its written permission.
Which models we use
We choose the provider based on the project. We work with OpenAI and Anthropic and, when the client asks for it, with local or private models running on the client's own servers.
WhatsApp only with consent
We only message people on WhatsApp who have given their consent, and always through the official WhatsApp Business API.
What we won't build
We turn down projects that involve:
- Spam or mass messaging to people who haven't given consent.
- Impersonating people or businesses.
- Making automated decisions with legal or similarly significant effects on someone without human review.
- Surveillance of people.
- Creating or spreading misleading content.
When the agent gets it wrong
AI can make mistakes. That's why we log errors, review them with the client and adjust the agent. If you saw one of our agents behave badly, email erik@taverassolutions.com.
Other legal pages