What it does ChatDBG wraps the standard Python, LLDB, and GDB debuggers with an OpenAI-powered assistant. When your C, C++, Python, or Rust program crashes, you type why at the debugger prompt. The tool feeds the stack trace and program state to GPT-4, then returns a root-cause analysis and a suggested fix. It can also answer follow-up questions and steer the debugger to gather more context.

The interesting bit The authors claim this is the first debugger to perform automatic root-cause analysis and suggest fixes. The integration is tighter than a side-chat window: the LLM can actually drive the debugger, running commands to inspect memory or variables to answer your question.

Key highlights

• Supports pdb (Python), lldb, and gdb (C/C++/Rust) with feature parity; other debuggers are being backported
• Works in post-mortem mode on core dumps, not just live crashes
• Includes Jupyter/IPython integration via %pdb magic and the ipyflow kernel
• Rust support via a #[chatdbg::main] attribute and panic = "abort" configuration
• Published as an FSE'25 paper with a full evaluation

Caveats

• Requires a paid OpenAI API key with a positive balance; GPT-4 access costs money per debugging session
• Native debugger setup is finicky: GDB needs Python 3.9+ support compiled in, and LLDB on Linux may require upgrading to LLVM 18
• The why command can fail silently if the extension didn’t load properly (the README includes a troubleshooting section for this)

Verdict Worth trying if you debug native crashes regularly and don’t mind paying OpenAI per incident. Skip it if you need offline-only tooling or if your GDB build lacks Python scripting support.