How AI Works

Bellwether uses Claude, an AI model built by Anthropic, to understand your questions and generate answers from your church data. Here’s what happens under the hood.

The Flow of a Question

When you type a question in Bellwether:

Your question goes to Bellwether’s server. It’s processed to understand what data you’re asking about.
Bellwether queries your ChMS. Based on what you asked, Bellwether makes API calls to Planning Center or Rock RMS to fetch the relevant records.
The data and your question are sent to Claude. The AI model reads the data and your question together to generate a response.
The response comes back to you. Bellwether displays the AI-generated answer in the chat.

What the AI Sees

When answering your question, Claude sees:

Your question and any follow-up context in the current conversation
The specific ChMS data fetched for that question (not your entire database)
Instructions about how to behave, format answers, and handle uncertainty

Claude does not have access to your full ChMS database. It only sees the specific records that Bellwether fetches for each question.

What the AI Doesn’t Do

It doesn’t train on your data. Anthropic’s API does not use your inputs or outputs for model training. Your church data doesn’t become part of the AI’s knowledge.
It doesn’t remember between sessions. Each conversation starts fresh. The AI doesn’t carry knowledge from previous conversations.
It doesn’t make things up silently. When Bellwether can’t find the data to answer a question, it says so rather than guessing.

Why Claude?

Bellwether uses Anthropic’s Claude model because of its strong performance on structured data questions, its clear communication style, and Anthropic’s data handling policies. Specifically, Anthropic’s API terms guarantee that customer data is not used for model training.