Skip to main content

LLM Usage Memorandum

Version 1.0 - 23 January 2026

Summary

Waterfox does not include large language models (LLMs) by default in the browser. The core experience remains focused on user control, clarity, and predictable behavior. If you choose to use external AI services, you may do so through extensions or your own setup.

Why this matters

“AI” is a broad term that covers many technologies. Waterfox draws a clear distinction between:

  • Constrained, single purpose AI tools (e.g., local translation models), and
  • General purpose LLMs that operate as opaque systems.

LLMs can be powerful, but they are black boxes - their behavior is difficult to audit, and their data handling can be unclear. Putting LLMs at the center of a browser risks shifting control away from the user.

Policy stance

Waterfox’s position on LLMs is guided by these principles:

  • User choice is core: No LLM features are enabled by default.
  • No opaque intermediaries: Browsing should be direct and user controlled.
  • Privacy first design: We avoid systems that require broad access to browsing context.
  • Transparency over convenience: Utility doesn’t override the need for explainable behavior.

What this does (and doesn’t) prohibit

This memorandum does not prevent users from:

  • Installing extensions that use external AI services.
  • Using AI tools in the browser on their own terms.
  • Re-enabling AI features that have been purposefully disabled.

It does mean that Waterfox will not:

  • Embed LLMs as a default browser layer.
  • Mediate user browsing through AI agents.
  • Introduce mandatory LLM features into core browsing.

Context and references

This memorandum aligns with the broader public position expressed in:

Interview excerpt

Q: You’ve been vocal about Waterfox staying “AI-free” while Mozilla is becoming an AI browser. I recently learned that Mozilla’s translation engine (that you referenced in your blog post) is not trained on open data, but from crawled copyright content. Does that change your perspective on inclusion within Waterfox?

A: I think “AI-free” isn’t my exact stance - “AI” is too broad a term to make blanket statements about. I’m cautious about LLMs specifically, and definitely against including them by default in the context of a web browser. If users want to use external AI services via extensions or their own setup, they’re free to do so - user choice is core to Waterfox’s ethos.

Q: Can you explain more about how you think about inclusion of AI features in Waterfox?

A: Outside of LLMs, ML has been a part of Waterfox for a while. Translations (Bergamot), TTS/STT (DeepSpeech), URL bar ranking for local history when entering keywords and more recently, alt text generation for PDF images. These all provide real utility to end users, are implemented transparently and have been a part of the browser for years. The biggest issue I find is with LLMs and how they are used, rather than the all encompassing “AI” term which includes many useful features over the years.

Changes to this memorandum

We may update this memorandum if our stance evolves or if the technology landscape materially changes. If that happens, we will document the changes and publish them clearly.