Think differently.

AI that reasons, not just responds.
Private. In your browser. No data leaves your device.

Scroll to explore
Philosophy

We believe AI should
listen before it speaks.

MentalChat is a fine-tuned language model built on LiquidAI's LFM2.5-350M architecture, trained specifically on mental health conversations. It doesn't diagnose — it reflects, paraphrases, and holds space.

Everything runs locally via ONNX Runtime and WebGPU. Your words never leave your browser. No servers. No logs. No surveillance. Just a conversation.

Method
01

Client-side
inference

The model runs entirely in your browser using WebGPU acceleration. Zero network calls after the initial load.

02

Quantized
precision

4-bit quantization keeps the model small enough for browser delivery while preserving the nuance of therapeutic language.

03

Empathic
fine-tuning

Trained on 16K curated mental health dialogues, emphasizing active listening, validation, and thoughtful reflection.

"The greatest compliment that was ever paid me was when one asked me what I thought, and attended to my answer."
Henry David Thoreau