VS-OPT Proposes Deterministic Pre-Query Verification for AI Search

Summary

The GitHub project VS-OPT and its GIUSTRA engine propose a deterministic pre-query pipeline for AI search and browser assistants. The project claims that current systems often rely on probabilistic, multi-turn refinement, increasing token use, latency, and operating costs, and says its approach can reduce LLM or search token overhead by 30% to 50%. Its pipeline moves from a known state to an event, a required control, and a newly verified known state. A rule-based query optimizer, designed to work with local Ollama deployments and Chromium sandboxes, corrects user intent before a search begins. GIUSTRA is intended to avoid repeated web crawling by checking source authority, timestamps, and versions at the point of use. The project also describes a zero-tracking fallback sandbox that maps intents deterministically and cites integration with a movie-search engine as an example. In its comparison, VS-OPT claims zero token waste in a single-pass intent flow, lighter client battery impact, and ledger-backed rather than probabilistically ranked data authority. The authors present it as a zero-training, zero-migration layer for privacy-focused Chromium browsers and distributed-state architectures. The article describes the architecture and claimed targets, but does not provide independent benchmark methodology or results beyond the stated comparison.