yuchuangu85/llm-anr

llm-anr is a Python tool that helps Android developers diagnose ANR errors, the popup that appears when an Android app freezes and the operating system decides it is no longer responding. Figuring out what caused an ANR usually means digging through large technical log files: thread traces, system event logs, memory snapshots, and kernel logs. This tool automates that evidence gathering step and then uses an AI model to suggest likely root causes. The workflow starts by pointing the tool at a bugreport, which is a diagnostic archive Android can export, a log directory, or a ZIP archive. The tool finds all relevant ANR evidence inside it, groups entries by individual ANR event so they do not mix together, then filters and parses each log type, including thread states, lock waits, input dispatch timeouts, and memory pressure signals. It packages the filtered evidence into a structured markdown file and works through a four phase analysis: examining the thread trace, the event log, the logcat, which is Android's general system log, and finally producing a combined conclusion. That last phase lists a timeline, the direct blocking point, candidate root cause chains, a confidence assessment, and remediation suggestions. The tool is deliberately cautious. It always marks results as requiring human confirmation and never claims a final verdict on its own. You can run each phase separately from the command line, use it interactively through an AI coding assistant such as Claude Code, or drive it directly through the Python API. The full README is longer than what was shown.