AI Model Disclosures

Effective June 1, 2026 · Version 1.0 · Operational reference; the Terms of Service and Privacy Policy control.

This page is the AI-model-version and material-change disclosure referenced by §P6.6 of the Privacy Policy (Appendix G). It identifies the artificial-intelligence (“ AI ”) and machine-learning (“ ML ”) models incorporated in the Terrestream Indoor Air Quality Sensor (the “ Device ”) and its associated Cloud Services at v4.0, together with capability and material-limitation statements for each. The Terms of Service and Privacy Policy control; this page is operational reference material only.

⚠️ Not a warranty; AI Outputs are informational only

The information on this page describes capability; it does not warrant performance. AI Outputs are probabilistic in nature and are subject in full to the limitations identified at limitations below and to the disclaimer and limitation-of-liability provisions of the Terms of Service (including, without limitation, §5, §7, §9, and §15). Reproduction or description of an AI model on this page is for User reference and does not constitute a warranty of correctness, completeness, diagnostic accuracy, or fitness for any particular purpose.

AI / ML models in production at v4.0

Discovery insights model

Function: Generates the weekly personalized Discovery cards described in the Free Tier feature matrix. Identifies patterns in the User’s indoor environment (e.g., “your home is calmest at 4-6 am; Saturday mornings are your worst”).
Tier: Free Tier (auto-activated).
Architecture: A cloud-hosted per-user statistical / ML pipeline trained on the User’s own Telemetry Data. Aerodyne may supplement per-user training with aggregate or deidentified cross-user signal in accordance with §P6.2 and §P6.3 of the Privacy Policy (population baselines, seasonal norms, and similar non-identifying inputs). Output phrasing is rendered by a third-party large-language-model (“ LLM ”) API; the structured insight is computed by the per-user pipeline and the LLM is used solely to phrase the resulting observation.
Inputs: The User’s Telemetry Data per §P3; aggregate population baselines per §P6.2. Personal Information beyond what is necessary to generate the insight is not included in the LLM prompt.
Outputs: A short observational text card surfaced in the Discover tab. Discovery cards do not push notifications by default (tier-features §).
Material limitations: See § Limitations below.

Annual climate letter and Wrapped narrative generation

Function: Generates the annual climate letter described in the Free Tier feature matrix and the multi-year “Wrapped” narratives described in the Pro Subscription feature matrix.
Tier: Free Tier (annual climate letter); Pro Subscription (multi-year Wrapped).
Architecture: Aerodyne calls a third-party LLM API to generate the narrative prose, with the User’s aggregated annual statistics as input. The current set of LLM providers Aerodyne may use is published, and updated as it changes, at /legal/sub-processors. Aerodyne does not commit to a specific provider for any given letter; the provider may be selected on a per-request basis from the published list.
Inputs: Aggregated per-user annual statistics derived from the User’s Telemetry Data per §P3. Personal Information beyond what is necessary to generate the letter (such as the User’s name, address, or contact information) is not included in the LLM prompt.
Outputs: A natural-language narrative surfaced in the Climate tab and (for Wrapped) at the corresponding seasonal moment.
Material limitations: See § Limitations below. LLM output is editorially shaped against the voice-and-tone constraints of §13 and may include factual inaccuracies inherent to LLM generation; Aerodyne does not warrant the factual content of an AI-generated letter.

Material limitations — what AI Outputs are NOT

The following limitations apply to every AI Output produced by any of the models identified above:

Not medical advice; not diagnostic. AI Outputs do not constitute medical advice, medical diagnosis, medical treatment, or a substitute for the judgment of a physician, nurse, allergist, public-health authority, or other qualified healthcare professional. AI Outputs do not diagnose disease, do not quantify health risk, and do not replace a clinical or professional opinion. See §7 of the Terms of Service.
Not for consequential decisions. AI Outputs must not be used as the sole basis for any consequential decision, including any decision related to housing, employment, insurance, education, child or dependent supervision, medication, evacuation, or the engagement of emergency services. A User who acts on an AI Output does so on the User’s own judgment.
No CO, smoke, fire, or life-safety inference. The Device does not measure carbon monoxide (CO), smoke, combustion gas other than CO₂, radon, or refrigerant; AI Outputs do not infer the presence or absence of any such hazard. A normal AI Output does not mean a CO, smoke, fire, gas-leak, or other life-safety hazard is absent. See the life-safety disclaimer at /legal/summary and §7 of the Terms of Service.
Probabilistic; phrasing does not imply confidence. AI Outputs are probabilistic and the phrasing produced by an LLM or other generative component does not imply quantitative confidence beyond what the underlying signal supports. Assertive or fluent phrasing is a property of language models, not of the underlying measurement. See §5 of the Terms of Service.
No unqualified safety language. Per §4.7 of the Terms of Service, AI Outputs do not use unqualified safety language (such as “safe,” “healthy,” or “all clear”) without an accompanying informational-only qualifier.

Opting out of AI features

As set out in §P6.5 of the Privacy Policy, You may opt out of AI features. Opting out disables AI Outputs (Discovery cards, climate letters, and similar surfaces); it does not disable core monitoring functionality. The opt-out control is exposed in the Mobile App and Web Dashboard settings.

Material changes and notice mechanism

Per §P6.6, Aerodyne documents material AI model versions and material changes on this page. A “ material change ” means the introduction of a new AI/ML model in production on a public Terrestream surface, the discontinuation of an existing production model, a change in the third-party LLM provider set used for AI Output phrasing, or a material change in the inputs used to train or query a model that bears on a User’s reasonable expectations under the Privacy Policy. For material changes, Aerodyne provides in-app notification per §P6.6. Non-material adjustments (such as routine retraining cadence, minor prompt revisions, or operational tuning that does not affect User-facing behavior) are not separately announced.

Version history

June 1, 2026 · v1.0. Initial publication. Models in production: Discovery insights model; annual climate letter / Wrapped narrative generation.

Prior versions are preserved at /legal/archive.

Cross-references

Privacy Policy §P6 — AI, training data, opt-out, model updates (the binding clause set for AI features).
Terms of Service §5 — AI, algorithmic outputs, and derived metrics (probabilistic-nature clause).
Terms of Service §4.7 — Product safety warning program (no-unqualified-safety-language rule for AI Outputs).
/legal/sub-processors — current set of third-party processors, including LLM providers used for AI Output generation.
/legal/tier-features — tier matrix; identifies which AI features are Free vs Pro.