Last updated: 2026-04-27
StelaHome uses AI assistance in several places: report-narrative generation, repair-guide drafting and refinement, troubleshooting question-and-answer features, quote-comparison summaries, and certain editorial and moderation workflows. Where AI is used, we disclose it on the relevant page or in the relevant section header.
AI processing today is performed through Anthropic Claude models served via AWS Bedrock and similar enterprise endpoints. All inference is server-side. No AI provider keys are exposed to your browser, and no client-side calls to a model provider are made from StelaHome pages.
AI inference receives the minimum data required to perform its task. For example: a report narrative receives anonymized property metadata and grade inputs; a Stela Fix guide-drafting prompt receives the creator’s authored content; a Quote Upload comparison receives OCR’d text from the uploaded contractor quote, after the text has passed our prompt-injection classifier.
Personally identifiable information — your full name, email, phone number, billing details, and unhashed home address — is not sent to AI inference endpoints. Where address-derived data is required for a downstream lookup, only an SHA-256 address hash or property coordinates leave the trust boundary.
AI-generated narrative passes through internal validators that suppress promotional language, steering language, and protected-class references before the text is rendered to you. We monitor for hallucinations and we invest in prompt-regression testing, but we do not guarantee that AI-generated text is free of error.
Stela Fix guide content goes through a multi-stage review pipeline that combines AI safety screening, peer review, and editorial review by our safety team. Composite-grade methodologies are not authored by AI — they are documented, versioned, and gated under our internal HD-18 governance regime.
You may report AI-generated content that is inaccurate, unsafe, or otherwise problematic to support@stelahome.com. We investigate reports, correct content where verified, and refine prompts and validators when a recurring failure mode is identified.
We do not knowingly use your account data, your home address, your uploaded documents, or your private session content to train third-party AI models. Where we use AI providers, we rely on the enterprise terms of service that prohibit using customer data for model training without consent.
We monitor state and federal AI-disclosure regimes — including California AB 2013, Colorado SB 24-205, and similar developing rules — and we update this disclosure as legal requirements evolve. If a specific feature triggers a stricter disclosure regime, we will surface the additional notice on the relevant page.
AI assistance does not replace professional advice and does not change the limits described in the Accuracy Disclosure. You remain responsible for independently verifying any fact you intend to rely on.