How Daylogue Uses AI

Daylogue uses AI in several parts of the product. This page explains exactly what the AI does, what it does not do, what data it sees, and how that data is handled at each step. No jargon, no vague reassurances. Just how it actually works.

What the AI does

AI powers five specific features in Daylogue:

• Follow-up questions during check-ins. When you check in, the AI reads what you wrote and generates a relevant follow-up question. This is what makes the check-in feel like a conversation instead of a form.
• Mood, energy, and stress extraction. The AI reads your words and identifies signals about how you are feeling. If you say "I barely slept and everything felt hard today," the AI notes low energy and elevated stress without you needing to move a slider.
• Pattern identification. Across days and weeks, the AI looks for recurring themes, correlations between metrics, and shifts in your emotional patterns. This is how Daylogue surfaces observations like "your stress tends to be higher on days when your sleep is below six hours."
• Narrative summaries. The AI reads across multiple check-ins and synthesizes them into a narrative about your recent days. This is not a summary of one entry. It is a synthesis of patterns across time, written in a reflective and warm tone.
• Chromascape color palettes. Based on the emotional tone of your check-in, the AI generates a unique color palette that represents the feeling of your day. This is an expressive feature, not an analytical one.

What the AI does not do

These are firm boundaries, not aspirational goals:

• It does not diagnose anything. The AI never labels you with a condition, suggests you might have one, or uses clinical terminology.
• It does not give advice. The AI observes patterns and asks questions. It does not tell you what to do. "Your energy tends to dip on days you skip exercise" is an observation. "You should exercise more" is advice. Daylogue does the first.
• It does not share your data. Your entries are not accessible to anyone but you. The AI processes them and returns results. It does not send them anywhere else.
• It does not remember you between sessions. Each check-in is processed with limited context. The AI does not build a persistent profile of you. It receives the current check-in, recent metric history, and your Focus Areas. That is all.
• It does not make clinical determinations. The AI is explicitly prompted to stay within wellness boundaries. It cannot and does not provide anything that could be mistaken for professional guidance.

How the processing works, step by step

Here is what happens when you complete a check-in that uses AI features:

1. 1.Encryption on your device. Your entry is encrypted with AES-256-GCM on your device before it is sent anywhere. The encryption keys are generated and stored on your device.
2. 2.Brief decryption for AI processing. For features that require AI (follow-up questions, pattern detection, narratives), the entry is briefly decrypted in memory and sent to the AI provider (AWS Bedrock) over an encrypted connection (TLS 1.3).
3. 3.AI processes and returns results. The AI reads your entry, performs the requested analysis, and returns the results. Raw entries are not stored or logged by the AI provider.
4. 4.Results are re-encrypted. The AI-generated results (pattern observations, narrative text, extracted metrics) are encrypted before being stored in the database.

What data the AI sees

During processing, the AI receives:

• The text of your current check-in (during processing only)
• Your recent mood, energy, and stress history (for context when generating patterns)
• Your Focus Areas (for relevance in follow-up questions and narrative themes)

The AI does not see your name, email address, or other personal identifiers. Your check-in content is not linked to your identity at the AI provider level.

AI training: a firm policy

Your entries are never used to train AI models. This is not a toggleable setting that could be changed in a future update. It is a policy decision that is also enforced contractually with the AI provider.

AWS Bedrock, the AI provider Daylogue uses, does not use customer data for model training by default. This is a reason Daylogue chose it.

The voice processing path

Voice check-ins use a separate processing path from text check-ins. Here is how it works:

• Audio is streamed to ElevenLabs in real-time for voice interaction. It is processed as it arrives, not stored as a recording.
• Transcription happens within the live session. After the session ends, neither the audio nor the transcription is retained by the voice provider.
• The resulting check-in text (what you said, transcribed) follows the same encryption and AI processing path as a typed check-in.