NovAST (github.com)

NovAST is an innovative middleware tool that significantly optimizes how developers interact with large language models (LLMs) by utilizing Rust and Tree-sitter to parse Abstract Syntax Trees (ASTs) from multiple programming languages, including TypeScript, JavaScript, Python, Java, and Dart. It drastically reduces LLM input tokens by up to 90% while preserving full architectural fidelity, enhancing context understanding without overwhelming the model with unnecessary details. Key features include a Cross-File Dependency Resolver for comprehensive context retrieval and an Autonomous Self-Healing loop that preemptively addresses syntax errors, ensuring smooth LLM integration. This tool's significance lies in its ability to streamline the coding process, allowing developers to generate precise and relevant code snippets efficiently. By focusing only on critical structural elements and eliminating irrelevant information, NovAST keeps LLM performance sharp and focused, mitigating issues like token explosion and context dilution common in traditional approaches. With future developments, including a fully Rust-based Language Server Protocol (LSP) Daemon, NovAST aims to become a cornerstone in autonomous, agentic coding tools. It invites contributions from the open-source community, empowering developers to shape the tool further.