What question did this study set out to answer?

The research aims to develop a privacy-preserving web automation system using a multi-agent approach.

April 26, 2026Open Access

Cosmos: A Privacy-Preserving Multi-Agent Framework for Browser-Native AI Web Automation

AAAayush aayushsoamDr. K.N.Modi University

Key Points

The research aims to develop a privacy-preserving web automation system using a multi-agent approach.
Developed a browser-native multi-agent AI framework combining planning and execution agents.
Evaluated performance against single-agent baselines through empirical testing and an ablation study.
Compared cost efficiency against commercial solutions across various large language model providers.
Achieved automation success rates of 95%, 92%, 88%, and 85%.
Demonstrated a 90–95% cost reduction compared to existing commercial solutions.
Reported an +11.2% improvement in performance over single-agent models (p < 0.01).

Abstract

Cosmos is a browser-native multi-agent AI framework for web automation, combining a Thinker (planning) and Navigator (execution) architecture. The system enables privacy-preserving, cost-efficient automation through a provider-agnostic LLM routing layer supporting OpenAI, Claude, Gemini, Groq, and local Ollama models. Empirical evaluation demonstrates success rates of 95%, 92%, 88%, and 85%, with 90–95% cost reduction compared to commercial solutions. An ablation study shows a +11.2% performance improvement over single-agent baselines (p < 0.01).

Ask AI

Helpful

Bookmark

View Full Paper