Back to services

Web-based AI

When we think about AI we immediately think about the powerful, large language models like GPT4, Claude, Mixtral or Llama. In the case of closed-source models, this requires relying on an expensive external provider. In the case of open source models, we can use an external providers or we can host the model ourselves on our own machines, typically in some cloud.

There is however also the possibility to run AI models directly in the browser. Nowadays, end-user devices like laptops or mobile phones increasingly ship with a GPU or even a dedicated AI-processor. WebGPU, the successor of WebGL, was designed with Machine-Learning in mind and is getting more and more supported, just recently on Android mobile.

Why web

While it is not possible to run very large models, smaller models can still be sufficient for many use-cases. Running AI models directly in the browser offers multiple advantages:

  • Lower latency since there is no network overhead. This allows for use-cases where the user expects real-time feedback.
  • Enhanced privacy since no data is transferred to a third party service.
  • Low cost since the model runs on the users' hardware.
  • Increased reliability since no third-party service is involved (see e.g. recent OpenAI outages).

How we can help you

At betalyra we have extensive experience with browser-based AI and can support you in the end to end design of your AI use-case:

  • UI/UX design of your use-case
  • Development in Typescript or Rust/WASM
  • Fine-tuning