ai_provider_browser

The Browser Provider is an experimental AI provider for Drupal that enables local, in-browser AI through Google Chrome’s Prompt API. Instead of connecting to a remote LLM service, this provider uses the browser’s built-in model running directly on the client’s machine.

Overview

This provider integrates Chrome’s Prompt API into the Drupal AI ecosystem. It allows Drupal sites to run any of the 100+ AI features using the user's own Chrome session and hardware, keeping data on the device and reducing external dependencies.

It is intended for early adopters, experimentation, and proof-of-concept work exploring browser-native AI before AI core 2.0.

Features

• Integrates Google Chrome’s Prompt API as an AI backend for Drupal.
• Executes prompts locally in the user’s browser session using an on-device model.
• Works with the Drupal AI ecosystem (AI, AI Agents, and related modules).
• Ideal for experimentation, prototyping, and research into browser-native AI workflows.

Requirements & Limitations

• Runs entirely on the client’s own computer in their Chrome browser. If you have background automation, the browser has to stay open!
• Requires Google Chrome with the Prompt API enabled via experimental flags.
• Requires at least 16GB of RAM on the client machine.
• The model is tied to the user’s Drupal account or session.
• Dependent on experimental Chrome features, which may change or break at any time.

Setup

To use the Browser Provider, the client must first enable Chrome’s Prompt API.

Open Chrome and navigate to chrome://flags and search for and enable the following flags:
#optimization-guide-on-device-model -> Set to "Enabled BypassPerfRequirement"
#prompt-api-for-gemini-nano -> Set to "Enabled"

Once the Prompt API is enabled in Chrome, the ai_provider_browser module can use it as a model within the Drupal AI framework. You will be prompted if you do not have a model, to download the model.

Status

The Browser Provider is highly experimental and should currently be treated as a thought experiment and proof of concept rather than a production-ready solution. You can theoretically use it for in house editors on Chrome, to save money.

As long as Chrome’s Prompt API and related browser AI features remain unstable or behind flags, this module will also be considered experimental and subject to change.

Roadmap

We want to add:

• Streaming support
• Better error handling
• WebAssembly support and support for other browsers

Note though that these will most likely be a natural part of AI 2.x or 3.x branches, rather then a external provider like this. This is mostly for testing in 1.x branch.