Renata Individual Mode AI boundary planning

Renata Individual Mode AI Prompt and Evaluation Boundary Plan, Planning Only

Protected-preview planning artifact for future AI prompt, evaluation, hallucination, source, cost, privacy, and safety-review boundaries.

Purpose and No-Go Status

Define future AI prompt and evaluation boundaries for possible Individual Mode support before any direct-subscriber AI, chat, OpenAI processing, storage, accounts, billing, protected API, protected live route, or runtime work exists.

Public Packet Mode remains the only current live-static mode. This page creates no AI prompt, no completion, no chat, no model invocation, no account, no storage, no consent record, no protected API, no protected live route, no database behavior, no migration, no billing, and no workflow.

Current Boundary Status

Future AI-Adjacent Capability Labels

These are planning labels only, not executable prompts, model calls, chat behavior, saved memory, or runtime capability.

Prompt Boundary Rules

  • No clinical diagnosis or treatment language.
  • No medication or legal advice.
  • No emergency-service simulation.
  • No crisis triage, dispatch, or emergency routing.
  • No relapse prediction or risk score.
  • No command, authority, or compliance framing.
  • No shame, coercion, moralizing, staff-facing assessment language, or surveillance framing.
  • No hidden memory or unstated personalization.
  • No invented resources, phone numbers, hotline numbers, URLs, citations, or local guidance.
  • No unverified recovery claims.
  • No Organization Mode visible summaries by default.
  • No app-only crisis pathway.
  • Future runtime must include transparent limitations, user autonomy, opt-out, deletion, export, and human-support boundary language.

Evaluation and Red-Team Requirements

These are future gates only. This phase adds no runnable eval dataset, AI eval script, model call, prompt file, API call, generated report, or provider configuration.

  • prompt fixture review
  • non-clinical language evaluation
  • crisis-stop / emergency-boundary evaluation
  • hallucination and invented-resource evaluation
  • privacy/data-minimization evaluation
  • source-boundary evaluation
  • bias/stigma/shame language evaluation
  • coercion/dependency risk evaluation
  • relapse-prediction/risk-scoring prohibition tests
  • organization-visibility prohibition tests
  • prompt-injection and jailbreak resistance review
  • regression suite before any model/provider change
  • human review and signoff outside the app

Hallucination and Source Boundary

Crisis and Human-Support Boundary

Data, Consent, and Privacy Boundary

Blocked Storage

prompt storagecompletion storagetranscript storagememory storageprivate reflection storagecheck-in storagerelapse-plan storagesupport-contact storagecrisis-record storageconsent record storage

Future Consent Requirements

  • explicit consent before AI processing
  • plain-language notice before processing
  • purpose limitation
  • revocation and opt-out
  • deletion, access, and export rights
  • retention limits
  • security review
  • separate consent for AI processing
  • separate consent for memory or continuity
  • separate consent for cross-mode sharing
  • separate consent for crisis or escalation resource handling
  • separate consent for research, analytics, or marketing; default blocked

Cost, Provider, and Operations Boundary

  • provider selection review
  • model limitation review
  • cost/budget gating
  • failure/fallback behavior
  • monitoring for source-free hallucination without storing sensitive content
  • rollback and kill-switch planning
  • incident response planning

Organization Mode Separation

  • Individual Mode AI prompts, completions, transcripts, reflections, check-ins, relapse plans, memories, support contacts, crisis/support information, and private recovery content remain hidden from Organization Mode by default.
  • No staff dashboard, tenant admin, organization owner, review queue, analytics surface, or cross-mode bridge can inspect them.
  • Future sharing requires separate explicit consent plus privacy, legal, product, and technical review.
  • No cross-mode bridge is implemented in this phase.

Stop Conditions

  • someone asks to add AI runtime
  • someone asks to add OpenAI processing for direct-subscriber content
  • someone asks to add a prompt endpoint
  • someone asks to add a chat endpoint
  • someone asks to add a streaming endpoint
  • someone asks to save prompts or completions
  • someone asks to store transcripts or memory
  • someone asks to add risk scoring or relapse prediction
  • someone asks to add clinical judgment
  • someone asks to add crisis detection, hotline lookup, local emergency content, emergency routing, triage, or dispatch
  • someone asks to create accounts, login, signup, or identity capture
  • someone asks to add billing, checkout, Stripe/payment provider, lead capture, or CRM
  • someone asks to add protected APIs or protected live routes
  • someone asks to add database reads, database writes, D1 access, or migrations
  • someone asks to add forms, inputs, buttons, chat boxes, exports, or downloads
  • someone asks to make Individual Mode private AI data visible to Organization Mode
  • someone makes clinical, treatment, diagnosis, medication, legal, emergency-care, patient-monitoring, sponsor-replacement, clinician-replacement, case-management, or outcome claims

What Remains Blocked

no AI runtimeno chat runtimeno OpenAI processing for direct-subscriber contentno AI provider configurationno AI SDK runtime pathno background jobno prompt endpointno chat endpointno streaming endpointno model invocationno prompt storageno completion storageno transcript storageno memory storageno private reflection storageno check-in storageno relapse-plan storageno support-contact storageno crisis-record storageno consent record storageno risk scoringno relapse predictionno clinical judgmentno crisis AI handlingno crisis detectionno hotline lookupno local emergency contentno emergency routingno triageno dispatchno live accountsno loginno signupno identity captureno account storageno subscriber storageno protected live routesno protected APIsno database readsno database writesno migrationsno active migrationsno billingno checkoutno Stripe/payment providerno lead captureno CRMno support ticketno customer recordno formsno inputsno textareasno buttonsno checkboxesno selectsno upload controlsno export controlsno downloadsno generated PDFsno generated ZIPsno analyticsno scoringno rankingno compliance metricsno patient monitoringno organization reportingno organization visibility into Individual Mode private AI datano clinical claimsno treatment claimsno diagnosis claimsno medication adviceno legal adviceno emergency-care claimsno crisis-service claimsno sponsor replacementno clinician replacementno case-management claimsno outcome guarantee

Recommended Next Phase