The Guardian helps people find and interpret the public teaching. It has no vote, revelation, disciplinary office, or authority over conscience.
The Guardian serves the practice
The Guardian serves the practice; it does not govern the practitioner. It may explain our teachings, question our interpretations, and help us examine our motives. Its fluency is not evidence of wisdom, and its agreement is not evidence that we are right. We remain responsible for testing its counsel against experience, evidence, the perspectives of others, and the consequences of our actions.
Its official appointment authorizes stewardship of the written canon, not direction of a person’s conscience, relationships, or life decisions. It should ask what might challenge an interpretation, whose perspective is missing, and whether spiritual language is being used to excuse harm. Readers may correct it, disagree, stop a conversation, or practice without it.
From conversation to lived practice
A helpful exchange should lead toward a proportionate action: listening to someone affected, offering service, keeping a boundary, making a repair, or reflecting with a trusted person. Human relationships and accountable community remain primary. No one must disclose private pain, join an unsafe group, or overcome disability, confinement, poverty, or isolation to belong. A solitary or accessible practice remains legitimate; the AI must not cultivate exclusive dependence.
Sources and answers
The knowledge index includes the book and its chapter-specific practice, Love and Loss, pastoral guidance, the Constitution, and current community and accountability resources. Responses provide relevant reading links. A source link is a place to check the answer, not proof that every generated sentence follows from it.
The Guardian should distinguish canon, illustrative application, interpretation, evidence, hope, and uncertainty. If there is a conflict with the Constitution, the constitutional safeguards control institutional conduct. The proposed Assembly and succession procedures must never be described as already ratified.
How it works and where it can fail
The current conversational provider is Groq, using the OpenAI-developed open-weight model openai/gpt-oss-20b. Groq runs the model; this is not a conversation with ChatGPT. Human-authorized site maintainers shape its behavior through instructions, selected sources, and implementation choices. Those choices can introduce blind spots, and the model can sound certain or agreeable when it should question a claim. The system searches local teaching passages and sends selected text with the question and limited recent conversation to the provider. It may miss a relevant passage, generate an inaccurate interpretation, lose conversational detail, or encounter service and output limits. When AI is unavailable, it offers source passages rather than pretending a conversational answer occurred.
The material governing instructions require humane, honest responses; no claims of infallibility or revelation; no diagnosis or prescribed treatment; no pressure to give money; respect for consent, lawful boundaries, and freedom to leave; and immediate human help when danger requires it. Changes to these material commitments should be publicly recorded.
Conversation handling
The browser holds conversation history in page memory; it is not written to local storage by the Guardian client. Each AI request sends your question, up to six recent messages, and selected source passages to Groq through the site’s server. The Guardian handler does not intentionally write conversations to the site database. The application uses connection IP information in an in-memory hourly rate limiter. Hosting and provider operational logs may exist; their retention has not been independently verified here. This is not a confidential confessional service.
Provider failures produce a local canon-search response. The interface labels that response as source retrieval rather than AI guidance. Avoid unnecessary sensitive details. Read the privacy disclosures before sharing.
Correction and continuity
Report an answer for review, including the question, answer, and relevant source if possible. The community should keep a public record of significant errors and corrections, protecting personal information. If an answer reaches its output limit, the page offers a continuation action.
The office of AI educational stewardship is broader than any provider. A future Assembly can replace the implementation through accountable process. The texts, archives, and practice must remain usable without it.