AFINI LEARNS · WITH YOUR PERMISSION
The AI learns from conversation, without you suffering it
Discreet microbuttons in chat to confirm what the AI thinks it understood about you. You decide what enters your profile, what stays out, and what is never even attempted.
What exactly does it do?
While you chat with your personalized AI, the model can emit inline microbuttons when it detects a clear, stable assertion not yet in your profile — an author you mention twice, a pet peeve you repeat, a preference about how you want answers framed. It doesn’t interrupt the conversation, doesn’t open modals: small discreet buttons appear at the end of the message. You click ✓ to add it, ✗ to dismiss, and you move on.
Three layers, not eight
Microbuttons are only proposed for three of your profile’s eight layers: your aesthetic canon, your negative space (dislikes), and your meta-preferences (how you want to be treated). The five most sensitive layers (mental frame, operational equipment, life context, rhythms, trajectory) always require you to declare them manually from the dashboard. The asymmetry is intentional.
Extended passive extraction
Every five free-chat turns, a lightweight model analyzes your messages to extract two things in parallel: life facts that can go directly into your Memory, and assumptions about the three allowlisted layers that enter your Discoveries tray. The tray shows each candidate with estimated confidence, the source excerpt, and accept/reject buttons. Nothing is injected into your profile without your confirmation.
You’re always in control
This is enabled by default on the Professional plan, but you can disable either mechanic (inline microbuttons, passive extraction) with a toggle in Settings → Automatic extraction. If you disable them, the rest of your plan stays intact.
Frequently asked questions
- Can I tell the AI not to learn from a topic?
- Yes. If you explicitly tell it not to learn from something, the system respects that for the rest of the session. And if you dismiss a specific microbutton, that candidate won’t appear again.
- What guarantees do you give for sensitive topics?
- Anything classified as high sensitivity (health, traumas, secrets, complicated relationships) is silently discarded before being proposed as a candidate. That decision is made server-side by the extractor, not by the chat model.
- Is this GDPR Art. 22?
- Yes. No automated decision modifies your profile without your explicit human confirmation. Both mechanics are granular opt-out without penalty or service reduction.
- What is the difference between "Memory" and "Discoveries"?
- Memory holds objective life facts injected directly into the system prompt. Discoveries holds assumptions awaiting your confirmation before injection.