AI Transparency
Last updated April 2026
Custodia uses AI sparingly and transparently. This page describes where AI is used, what models are involved, and what guardrails we apply.
Where we use AI
- Weekly hero summaries: a short narrative drawn from your week's signals (volunteer hours, observations, monitoring entries).
- Reflection prompts: contemplative prompts based on your organization's focus areas.
- Library recommendations: matching public-domain works to your stewardship themes.
Models and routing
AI requests are routed through the Lovable AI Gateway to vetted providers (Google Gemini, OpenAI). The specific model is chosen per task to balance quality, cost, and latency.
What we do not do
- We do not train third-party AI models on your stewardship data.
- We do not use AI to make consequential decisions on your behalf (grant outcomes, volunteer eligibility, ecological judgments).
- We do not generate fabricated species observations, monitoring entries, or volunteer hours.
Human in the loop
AI outputs are presented as drafts and prompts, never as facts. Stewards review and edit before any AI-assisted text becomes part of the record.
Opting out
You may disable AI features for your organization at any time from Settings. Existing AI-generated content is retained as part of your historical record unless you delete it.