Orbital Collaborative Computing Between General-Purpose Internet Search Engines and Large Language Models: A Novel Framework for Bidirectional Search–Generation Integration in Space

Conventional Internet search engines and large language models (LLMs) have predominantly evolved as separate computational systems, with their integration generally relying on terrestrial cloud infrastructure and ground-based data centers. Meanwhile, emerging space-based computing technologies have created new opportunities for deploying increasingly sophisticated computational services in Earth orbit. This paper proposes Orbital Collaborative Search–LLM Computing (OCSLC), a novel framework for deploying general-purpose Internet search engine infrastructure and large language models within a unified orbital computing environment and enabling bidirectional, iterative collaboration between these two systems. Unlike conventional satellite-assisted information retrieval architectures, domain-specific satellite data search systems, and terrestrial retrieval-augmented generation (RAG) applications, the proposed framework specifically investigates the operation of a general-purpose Internet search engine in orbit and its direct computational collaboration with an orbital LLM. Under the proposed architecture, users on Earth submit natural-language requests through available communication networks to an orbital computing platform. The orbital search engine processes search requests through query interpretation, index access, document retrieval, relevance evaluation, and ranking. Retrieved information is subsequently passed to the orbital LLM for contextual understanding, reasoning, synthesis, and response generation. Importantly, the collaboration is not restricted to a conventional one-directional retrieval-to-generation pipeline. The LLM can dynamically formulate additional search queries, request further retrieval operations, identify missing or conflicting information, evaluate intermediate results, and initiate repeated search–analysis–generation cycles. Conversely, search operations can be initiated or refined in response to the LLM's intermediate reasoning and information requirements. This creates a bidirectional computational relationship in which information retrieval and generative reasoning continuously influence one another. The framework further considers the deployment of orbital web-indexing infrastructure, distributed search storage, LLM inference services, search–generation orchestration, inter-satellite communication, and satellite-ground interfaces. Multiple orbital computing nodes may cooperate to distribute indexing, retrieval, and inference workloads, while terrestrial gateways facilitate access to Internet content and communication with external users. The paper also investigates potential mechanisms for maintaining search-index freshness, coordinating retrieval and inference resources, and managing computational tasks under orbital communication and energy constraints. Unlike existing research that primarily investigates satellite communication optimization, onboard artificial intelligence, domain-specific satellite information retrieval, or conventional RAG systems, this study focuses on the integration of general-purpose Internet search infrastructure and generative language intelligence as mutually coordinated services within an orbital computing environment. To the best of our knowledge, the specific combination of an orbital general-purpose Internet search engine and an orbital LLM operating through a bidirectional and iterative search–generation architecture represents a distinct research direction not identified in the literature examined for this study. This paper establishes its conceptual architecture, operational mechanisms, research questions, and principal implementation challenges, providing a foundation for future investigation into intelligent space-based Internet information services.