AI Safety
AI Readiness
WarpIntel can prepare AI-assisted analysis without making live provider calls by default. Public tools keep deterministic local results ready while OpenAI usage stays behind protected setup review, explicit feature flags, approved WarpIntel/Olympus key ownership, and no-secret smoke checks.
Public AI readiness is count-only; provider setup details remain protected.
Initial analyzer workflows are prepared for deterministic-first AI enhancement.
Setup packet, dry-run, and activation rehearsal coverage stay behind admin-only routes.
Owner review evidence can be packaged without OpenAI calls or token spend.
Readiness dry runs do not call OpenAI, submit prompts, or spend tokens.
WarpIntel/Olympus account boundary
Protected setup guidance requires a WarpIntel/Olympus OpenAI project boundary and excludes separate-project account crossover.
Feature flag and key gate
AI enhancement remains inactive unless the app feature gate and approved provider key presence are both ready.
Spend and budget review gate
Production AI spend requires owner approval before first-tool AI calls are allowed.
First-tool scope boundary
Initial AI eligibility is limited to Intel Desk, Fit Coach, and Loss Reviewer while other tools stay deterministic.
Deterministic fallback required
Public tools must keep returning deterministic local analysis if AI is disabled, unavailable, or rejected for a request.
Protected dry run without provider call
Protected dry runs validate readiness without calling OpenAI, submitting prompts, or spending tokens.
Input and output safety boundary
Input size is capped and AI output remains advisory, separated from evidence and human-reviewed decisions.
First-tool smoke review gate
A non-sensitive first-tool smoke test must pass before live AI enhancement is considered active.
Eligible Tools
Intel Desk, Fit Coach, and Loss Reviewer can use deterministic analysis now and AI later.
Protected Checks
Setup packet, dry-run, and activation rehearsal stay behind protected admin routes.
Activation Rehearsal
Owner review can package first-tool AI activation evidence without provider calls or token spend.
Fallback
Public tools keep returning deterministic local analysis if AI is disabled or unavailable.
Provider Calls
Readiness checks do not call OpenAI, submit prompts, or spend tokens.
intel desk
Prepared for deterministic local analysis first; AI enhancement stays opt-in and feature gated.
fit coach
Prepared for deterministic local analysis first; AI enhancement stays opt-in and feature gated.
loss reviewer
Prepared for deterministic local analysis first; AI enhancement stays opt-in and feature gated.
Provider Keys
Excluded from public pages, health snapshots, smoke logs, screenshots, and readiness feeds.
Model Env Values
Excluded from public pages, health snapshots, smoke logs, screenshots, and readiness feeds.
Billing Details
Excluded from public pages, health snapshots, smoke logs, screenshots, and readiness feeds.
Raw Prompts
Excluded from public pages, health snapshots, smoke logs, screenshots, and readiness feeds.
Raw Pasted Inputs
Excluded from public pages, health snapshots, smoke logs, screenshots, and readiness feeds.
Token Spend
Excluded from public pages, health snapshots, smoke logs, screenshots, and readiness feeds.
Private Eve Data
Excluded from public pages, health snapshots, smoke logs, screenshots, and readiness feeds.
Protected Setup Details
Excluded from public pages, health snapshots, smoke logs, screenshots, and readiness feeds.
No-Secret AI Rule
Public AI readiness exposes aggregate first-tool counts, feature-gate booleans, activation criteria, activation-rehearsal counts, protected-check booleans, deterministic fallback status, and no-provider-call dry-run status only; it does not expose provider keys, model environment values, billing details, raw prompts, raw pasted analyzer inputs, EVE tokens, private account data, or token-spend records.