Kanha

Train AI chatbots on your website. Run them entirely on-device.

kanha.ai · Docs · npm

Kanha fine-tunes compact language models on your website's content and deploys them directly in the browser via WebGPU. No server-side inference. No per-query costs. No data leaves the device.

Our Mission

We're making intelligent chatbots accessible to every business — regardless of size or technical expertise — without the burden of recurring per-query costs or complex infrastructure.

What's in this org

Each repository here is a trained model checkpoint for a Kanha customer's bot. Models are quantized for efficient on-device inference and served to the kanha SDK at runtime.

These models are not general-purpose — each is custom-trained on a specific website's content and optimized to answer questions about that site.

How it works

1. Crawl your website (handles static and JavaScript-rendered pages)
2. Generate question-answer training data from your content
3. Fine-tune a compact model tailored to your site
4. Deploy with a single script tag — runs on-device via WebGPU

Who it's for

• Businesses looking to add intelligent support to their websites
• Companies aiming to eliminate recurring costs of cloud-based chat solutions
• Organizations that prioritize data privacy — all processing stays on-device
• Startups and enterprises seeking customizable, scalable chatbot solutions

Links

• Website
• Documentation
• SDK on npm
• Showcase
• GitHub