Abstract
A system and method for dynamic, query-guided synthesis and injection of a contextual session essence into large language model prompts provide a browser-native solution to computational latency and context window saturation. The system comprises a context gathering module that performs real-time state measurement of Document Object Model mutations and session logging, contingent upon a user-authorized processing flag. A context analysis and inference engine transforms the resulting raw data cluster into a high-density natural language summary, referred to as the contextual session essence, by executing on-device machine learning inference for intent classification and entity extraction. The query augmentation module intercepts user queries at the browser networking stack and merges the contextual session essence into the request payload. The described methodology addresses the engineering inefficiencies of conventional systems by reducing token consumption and optimizing the focus of the external large language model through the proactive removal of irrelevant session noise to reduce the computational burden on the external processing interface.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Prakash, Abhay and Chang, Sophie, "Browser-Integrated Bidding System For Third-Party AI Agent Tools", Technical Disclosure Commons, (January 22, 2026)
https://www.tdcommons.org/dpubs_series/9213