How Bella works and what data she uses

Bella is powered by a curated set of state-of-the-art AI models: ChatGPT 5.6 Luna (the no-cost Auto default) plus ChatGPT 5.6 Terra (OpenAI) and Claude Haiku 4.5, Claude Sonnet 5 and Claude Opus 5 (Anthropic). You can choose which model powers your conversations from the model selector in chat; the premium models are usage-based, billed from your monthly AI allowance or a purchased balance top-up.

We do not train or fine-tune these models, and none of our material goes into any training dataset. What makes Bella unique is not the underlying AI model but the additional layers built on top of it:

1. Human Design knowledge

Leading AI models already carry an extensive library of Human Design material: gate and line descriptions, channel dynamics, type strategies, authority types, profiles, variables, incarnation crosses, and teachings from Ra Uru Hu. Our specialty is drawing that knowledge out reliably — the context, structure and prompting we give each model — rather than a custom-trained dataset.

2. Your personal chart data

Bella receives your actual Bodygraph data with every message: all 1,000+ data points including activated gates and channels, defined and undefined centres, profile, type, authority, and more. This is what allows Bella to give personalised answers rather than generic Human Design explanations.

3. Chart context and memory

Bella can use Chart Context notes you add for any chart, plus Bella Memory (saved context from previous conversations), to further personalise responses over time.

Accuracy

With the above layers, Bella achieves around 98-99% accuracy when reading Bodygraph data. Accuracy is continuously improved by refining the system prompts and how charts and Human Design context are presented to each supported AI model — never through model training.