How we use AI

AI writes drafts; people decide

Where Mycelity uses AI, such as summarising survey answers, suggesting taglines and poster text, finding events for Discover or reading in a grant call, what it writes is a draft for the person who asked for it. They read it, change it or throw it away before relying on it.

AI makes no decisions about people. Shortlisting, acceptance, payments and certificates are decided by organisers and by fixed rules in our code. Numbers, scores and layouts are worked out by code, not by AI.

Where it runs

By default, AI runs on Mycelity's own servers, using open models (Llama, and models from Hugging Face). What it works on doesn't leave the platform.

What goes to Claude

Only if an organisation, or you, switch on Fast AI, a paid option. Then the text a feature works on is sent to Anthropic, the maker of Claude, to process, and the result comes back to the feature.

Answers to questions marked as identifying (names, emails, phone numbers) are never sent to any AI model. Features that would send other personal data stay off until a Platform Admin switches them on.

No training on your data

We don't use your data to train AI models. The models on our servers don't learn from what they process, and under Anthropic's commercial terms, what we send through its API isn't used to train its models.