I built a small tool to reduce input token costs by 20-30% for agentic tasks (bigindexer.com)

A developer has created an open-source tool called Big Indexer (BGI) aimed at reducing input token costs for AI-driven coding assistant tasks by 20-30%. This tool comes in response to challenges faced by users of AI coding assistants, particularly the unreliability of answers related to code architecture when relying on models that embed codebases based purely on textual similarity. BGI addresses this by constructing a behavioral graph of the code repository, which provides the AI with a more accurate understanding of the architectural context. The setup process is quick, requiring just a few installation commands. For the AI/ML community, BGI's significance lies in its ability to enhance the quality of AI-generated code suggestions by improving boundary accuracy and actionability metrics. In extensive tests with open-source repositories, BGI demonstrated a decrease in median agent latency from 133.8 seconds to 66.2 seconds and improved the models' ability to accurately locate relevant code structures, as highlighted by a dramatic improvement in performance on a complex codebase. The project underscores a shift towards compact, local solutions that operate without the costs associated with cloud-based services, offering developers a viable alternative in the current landscape of AI coding assistants.