Apps

Edge adds Aion Instruct for on-device AI in websites

· Geeknewz Author

A silver Surface Laptop 7 showing the Windows boot logo, resting on its box on a wooden table.
Photo: StrangeApparition2011 / Wikimedia Commons, CC0 1.0. July 11, 2024 archival photograph, proportionately resized. Illustrative Windows laptop; Aion and WebBrain are not shown. Source and dedication linked in Sources.

Microsoft Edge's new built-in AI model and Mozilla's latest extension spotlight point to the same practical question: where does the particular browser feature you're using send its work? The answer depends on the feature and its configuration, even when local AI is part of the pitch.

Edge is changing the model websites can use on compatible devices. WebBrain offers a browser assistant with several model options. The useful takeaway from both announcements is to check how the specific task is configured before sharing page content.

Edge changes the model behind local website features

In its November 19 Aion Instruct announcement, Microsoft says Aion replaces Phi-4-mini as Edge's built-in small language model. Edge downloads it when a site or feature first needs it, and only on suitable devices. Turning off on-device AI in Edge settings removes the model and prevents future downloads.

Microsoft also opened an origin trial for the Prompt API, allowing participating websites to test features using the model. That is a developer rollout, not a promise that every website now has a local assistant. The separate Decisions API remains an experimental preview in Edge Canary and Dev.

For readers, the significant detail is Microsoft's description of hybrid use: developers can combine on-device models with a configured cloud provider. The availability of a local model doesn't establish which route a particular site takes. Look for the site's explanation of that feature before treating the browser's model download as a privacy guarantee.

WebBrain makes the configuration question explicit

Mozilla's November 20 WebBrain developer spotlight describes an extension that can summarize pages and help with browser tasks using offline models. It also notes support for paid AI accounts. That makes the current product documentation worth reading alongside the profile's broad local-processing claims.

The WebBrain Firefox add-on listing is more specific: local inference can use a model running through LM Studio or Ollama, while cloud models use a connected provider. It says the selected AI service processes the context needed for a request, and some connected tools use online services. Its built-in model also has a Help Improve WebBrain setting enabled by default.

The listing distinguishes read-only Ask mode from Act mode for browser actions, and lists access to data for all websites among the extension's required permissions. Those are useful details to review before installation; this is a description of the documented options, not a hands-on endorsement.

When evaluating a browser AI feature, decide whether you want help reading a page or permission to act on it, then check the selected model, connected services and relevant settings. If you're evaluating a new setup, start with public material. Let the configuration you can verify guide what you share.