A recent Chinese developer post demonstrates how to combine the Nebula SDK with Tampermonkey (a popular user script manager) to create an LLM-powered agent that persists across web pages. The approach effectively gives the LLM a 'body' that can interact with any website the user visits, enabling context-aware assistance, data extraction, or automated actions. For indie hackers and browser extension developers, this pattern offers a lightweight alternative to building full browser extensions—leveraging existing SDKs and user scripts to prototype AI companions quickly. The technique is particularly relevant as the ecosystem for embedding LLMs into everyday browsing grows, with potential applications in productivity, research, and personal automation. While the original post is in Chinese, the concept is globally applicable and could inspire similar projects using other LLM APIs and user script frameworks.
A practical guide to creating persistent LLM agents in the browser using Nebula SDK and Tampermonkey, enabling AI companions that interact with any webpage.