How it works

The science behind emotion-aware AI.

AquaMind is built on a simple premise: how you say something matters as much as what you say. Here's the pipeline, the models, and the principles behind every reply.

The three-step pipeline

Every message goes through three lightweight passes before the model writes a single word back to you.

Step 1

Listen

A small classifier reads your message for sentiment (positive, neutral, negative) and tone (calm, anxious, excited, frustrated, curious, and a few more). It runs in under 50ms and never leaves your conversation.

Step 2

Frame

The detected signals are added to the prompt as context, alongside your conversation memory. The model knows you're stressed before it picks a single word — so it can lead with calm, not chirpy.

Step 3

Respond

The chosen model generates a reply tuned to your tone. We surface what we detected back to you — no hidden personalization — so you always know what the system saw.

The four models

One stack, four tiers. Each model is tuned for a different balance of speed, depth, and developer access.

Octana

The base augmented model.

Our entry-level model. Fast responses, full sentiment + tone awareness, and a 10-message daily limit on the free plan. Great for daily conversations and getting a feel for emotion-aware AI.

Sondet

Tuned for tone and nuance.

Same pipeline, deeper tone modeling — better at picking up sarcasm, hesitation, and mixed emotions. 500 messages a month on the Sondet plan.

Kodekai

Our most capable, long-context model.

Long-context reasoning with persistent memory across chats. Mood-adaptive replies and the strongest emotional consistency. Unlimited messages on the Kodekai plan.

Tharus

For builders.

Everything in Kodekai plus extended context, API access, the highest rate limits, and webhooks for emotion + tone signals so you can build on top of AquaMind.

The principles behind every reply

Emotion-aware AI can quickly become manipulative. These are the guardrails we hold ourselves to.

Transparency over personalization

We tell you what we detected in your tone, on every message. No silent profile, no behavioral targeting, no hidden flattery loop.

Augment, don't replace

You stay the central authority on your decisions. The model handles the heavy lifting; intuition, judgment, and context stay with you.

Calm by default

When in doubt, we default to clarity and calm — not energetic agreement. We'd rather be useful than entertaining.

Your data, your control

Conversations are private to your account. You can delete history any time, and we never train on your messages without explicit opt-in.

Try it for yourself.

Reading about emotion-aware AI only gets you so far. Send a message and feel the difference.

See pricing