AI & Automated Systems Notice

Version 1.0 | Effective: September 6, 2026

ALYT uses artificial intelligence to understand what you say and to help decide what your home should do about it. This page explains, without marketing language, what that AI actually is, where it runs, what it is good and bad at, and what we do and do not do with the data it processes.

1. Where AI Is Used in ALYT

Wake-word detection — a small model listening continuously for your hub's wake phrase (for example, "hey Simone")
Speech-to-text — converting your spoken words into text once the wake word is heard
Language understanding and response generation — interpreting what you asked for and producing a spoken reply, including deciding whether to run a routine, execute a built-in command, or generate a conversational answer
Text-to-speech — synthesizing the assistant's spoken reply

As of this notice's effective date, ALYT does not use computer vision or facial recognition on any camera feed — native camera support is still in development, and third-party camera integrations (like Ring) send their video through the third party's own systems, not ALYT's AI.

2. Local vs. Cloud Processing

Every stage listed above can run entirely on your ALYT hub, using models deployed to the device rather than called over the internet, and this is the default configuration. Only if your household explicitly opts in to "cloud reasoning" (Settings → Voice, off by default) can the language-understanding step be routed to Fizzify's own cloud inference service instead. This distinction, and exactly what stays local versus what can optionally leave your home, is described in full technical detail in our Privacy Policy, Section 3.

3. AI Is Probabilistic, and It Will Sometimes Be Wrong

ALYT's voice assistant is not, and does not claim to be, perfectly accurate. Like every AI system available today, it can misunderstand what you said, fail to recognize a valid command, or generate a confidently-stated answer that is simply incorrect.

Speech recognition limitations

Accuracy is affected by background noise, accents, unusual vocabulary, multiple people talking at once, and distance from the microphone. A misheard word can change the meaning of your entire request.

Language understanding limitations

The assistant may misinterpret an ambiguous request, fail to understand a command phrased in an unexpected way, or — when using a general-purpose language model for open-ended questions — occasionally produce an answer that sounds confident but is factually wrong (sometimes called a "hallucination" in AI research).

What this means for how you should use ALYT

Do not rely on the voice assistant, or any AI-driven automation, for time-critical or safety-critical actions — for example, do not build a routine where an AI-interpreted voice command is the only way to unlock a door during an emergency.
Verify important commands by checking the resulting device state, especially for actions with real-world physical consequences (locks, security-relevant switches, appliances).
Treat conversational answers from the assistant as helpful, not authoritative, particularly for anything factual, medical, legal, or safety-related.

4. AI-Driven Home Automation

Routines and automations you build can be triggered by voice commands interpreted by AI, but the routine itself — once triggered — executes a deterministic, pre-defined sequence of actions you configured; the AI's role is limited to recognizing that you asked for it, not to deciding what the routine does. This means a routine's behavior is exactly what you configured, every time, even though the trigger recognizing your request is probabilistic.

ALYT does not currently use AI to autonomously decide to take action without a trigger you defined (a routine, a schedule, or a direct voice command) — it does not, for example, use AI to infer that "something seems wrong" and take independent action. Any such capability, if built in the future, will be described here and will be opt-in.

5. No Guarantee of Security Event Detection

If you use sensor-based or (through an integration) camera-based automation for security purposes, understand that ALYT does not guarantee detection of any particular event — an intrusion, a fire, unusual motion. Sensor and connectivity failures, AI misclassification, and network outages can all cause a genuine event to go undetected or unreported. See our Terms of Service, Section 9 for the full statement on this.

6. Human Control Remains With You

You can always override, disable, or delete any AI-driven feature or automation. Nothing in ALYT is designed to resist your direct control of your own home — a manual device command always takes precedence, and you can turn off voice listening per room, disable an automation, or disconnect cloud reasoning at any time from Settings.

7. Is Your Data Used to Train AI Models?

No. We do not use your voice recordings, transcripts, commands, home configuration, or any other personal data collected through ALYT to train, fine-tune, or otherwise improve any AI model — ours or a third party's. The models ALYT uses are pre-trained and deployed as-is; they do not learn from your household's specific data while running in your home.

This is a direct statement of current practice, not a hedge. If this ever changes, we will update this notice and our Privacy Policy prominently, before the change takes effect, and we would seek your explicit, freely-given, opt-in consent before any such use — not treat it as already covered by a generic "may improve our services" clause.

8. Cloud Reasoning: What Happens to a Request

If your household has opted in to cloud reasoning, a query routed to the cloud is processed by Fizzify's own inference infrastructure and answered; we do not retain a running conversational history at the model layer beyond what is needed to correctly resolve an immediate, obvious follow-up within the same exchange (for example, correctly answering "and Friday?" right after a question about Thursday's weather). Whether a given voice turn used cloud processing at all is recorded in that turn's metadata, described in our Privacy Policy, Section 6, so you can verify it directly rather than take our word for it.

9. Vision and Computer Vision — Not Yet Built

ALYT's marketing describes an eventual goal of local, on-premises computer vision for camera-based automation (for example, distinguishing a person from a package on your doorstep). As of this notice's effective date, this capability does not yet exist in the shipping product. When it does, this notice will be updated with its specific accuracy limitations and data-handling details before it ships, not after.

10. Questions

If you have questions about how AI is used in a specific ALYT feature, or want more technical detail than this notice provides, email support@alyt.co.

Contact

Fizzify Inc.
30 N Gould St STE N
Sheridan, WY 82801
United States

privacy@alyt.co | support@alyt.co

Last updated: September 6, 2026