Nvidia's Hyperlink Agent Brings Private AI-Powered Search to Local PCs

Nvidia's AI tool finds everything you saved without leaking a byte of data

LLM inference on Hyperlink doubles in speed with Nvidia's latest optimization Nexa.ai's new "Hyperlink" agent introduces an approach to AI search which runs entirely on local hardware. Designed for Nvidia RTX AI PCs, the app functions as an on-device assistant that turns personal data into structured insight. Nvidia outlined how instead of sending queries to remote servers, it processes everything locally, offering both speed and privacy. Hyperlink has been benchmarked on an RTX 5090 system, where it reportedly delivers up to 3x faster indexing and 2x the large language model inference speed compared to earlier builds. These metrics suggest it can scan and organize thousands of files across a computer more efficiently than most existing AI tools. Hyperlink does not simply match search terms, as it interprets user intent by applying the reasoning capabilities of LLMs to local files, allowing it to locate relevant material even when file names are obscure or unrelated to the actual content. This shift from static keyword search to contextual comprehension aligns with the growing integration of generative AI into everyday productivity tools. The system can also connect related ideas from multiple documents, offering structured answers with clear references. Unlike most cloud-based assistants, Hyperlink keeps all user data on the device, so the files it scans, ranging from PDFs and slides to images, remain private, ensuring that no personal or confidential information leaves the computer. This model appeals to professionals handling sensitive data who still want the performance advantages of generative AI. Users gain access to fast contextual responses without the risk of data exposure that accompanies remote storage or processing. Nvidia's optimization for RTX hardware extends beyond search performance, as the company claims retrieval-augmented generation (RAG) now indexes dense data folders up to three times faster. A typical 1GB collection that once took nearly 15 minutes to process can now be indexed in about 5 minutes. The improvement in inference speed also means responses appear more quickly, making everyday tasks such as meeting preparation, study sessions, or report analysis smoother. Hyperlink merges convenience with control by combining local reasoning and GPU acceleration, making it a useful AI tool for people who want to keep their data private.

Meet Hyperlink: A local NotebookLM-style tool for you

Nvidia says local processing ensures sensitive materials never leave the machine making it suitable for professionals with confidential workloads. Nexa.ai has introduced a new AI search agent called "Hyperlink," designed to run entirely on local hardware. The application is built for Nvidia RTX AI PCs and functions as an on-device assistant capable of turning personal data into structured insights. By connecting information from notes, slides, PDFs, and images, the tool aims to organize and interpret user files effortlessly. Nvidia outlined that the key differentiator for Hyperlink is its ability to process queries locally rather than sending them to remote servers. This ensures that all scanned files and user data remain private on the device, preventing personal or confidential information from leaving the computer. This approach is intended to appeal to professionals who manage sensitive data but still require the performance benefits of generative AI. Performance benchmarks reported by TechRadar indicate significant speed improvements. When tested on an RTX 5090 system, Hyperlink delivered up to three times faster indexing and double the large language model (LLM) inference speed compared to previous builds. Nvidia claims that its optimization of retrieval-augmented generation (RAG) allows the tool to process dense data folders much more quickly; for example, a 1GB collection of data that previously took nearly 15 minutes to index can now be processed in approximately 5 minutes. Beyond raw speed, Hyperlink moves away from static keyword matching. It utilizes the reasoning capabilities of LLMs to interpret user intent, allowing it to locate relevant materials even if file names are obscure or unrelated to their content. The system can also connect related ideas across multiple documents to provide structured answers with clear references.