Nura and human authority

AI transparency

Nura is designed to be useful because she is embedded in the Care House—not because she is allowed to act without limits.

The live AI provider is not bundled with public credentials. The interactive demo uses deterministic local sample data and does not call an external model.

1. What Nura is

Nura is the living AI intelligence and guide inside the Care House. The implementation combines product context, deterministic handling, a Firebase semantic provider path, persona orchestration, source voice, guardian decisions and trusted action handoff.

2. What Nura may do

3. What Nura must not do

4. Source awareness

Source references are represented in Nura, structured drafts and document components. The persona layer distinguishes source present, missing and conflicting states. A source-aware answer may still be incomplete or wrong and should not be treated as professional verification.

5. Uncertainty and clarification

Domain models include clarification, source-required and permission-sensitive states. The guardian/persona policies can ask a question, block a claim when a source is missing, route to administrator or professional review, or use direct emergency wording.

6. Human approval and trusted commit

The app separates suggestions from committed care objects. Structured drafts and review items carry source references and review state. Nura’s command handoff opens existing task, appointment, note, document or review controllers rather than silently writing protected data.

7. Model provider and data flow

The Cloud Functions code supports a configurable external semantic provider using a protected endpoint and secret key. The archive does not prove which provider would be used in production, its hosting region, retention policy, training policy or contractual terms. The buyer must select and review the provider before deployment.

8. Evaluation and limitations

The project includes tests for source integrity, permission integrity, serious contexts, repetition and anti-flanderisation behaviour. Tests reduce risk but do not prove safety, accuracy or regulatory suitability in real-world care settings.

Contact: Acquisition and marketing-site privacy enquiries may be sent to pssmaysara@gmail.com.