APG: Agentic Static-Analysis Infrastructure on LLVM-IR Graph Databases
Story 7, from arXiv
APG is a new tool that enables LLM-based coding agents to perform static analysis on C/C++ code by leveraging LLVM-IR graph databases. It provides 38 atomic capabilities for analysis and repair, outperforming existing methods in benchmark tests.
Key points
- APG is built on SVF and LLVM, using LLVM-IR graph databases for static analysis.
- It provides 38 atomic capabilities for LLM agents to select, compose, and orchestrate.
- On benchmark repair tasks, 35 of these capabilities achieve over 50% first-attempt query success.
- APG improves first-iteration repair success by 17.6–18.0 percentage points over baselines.
- It reduces average LLM input-token consumption by 90.8%.
- The tool is available for use and includes a demonstration video.
Why it matters
As LLM-based coding agents become more prevalent, tools like APG offer a structured way to integrate static analysis into their workflows. This can improve the reliability and efficiency of code repair and analysis tasks, especially for C/C++ projects.
Counterpoint
APG's effectiveness depends on the quality of the LLVM-IR graph databases and the specific analysis tasks. It may not generalize well to other programming languages or more complex codebases without additional adaptation.
Sources
This summary was written by a language model and may contain errors. Read the original for the full story.