kong
View on GitHubThe world's first agentic reverse engineer.
Kong is an agentic reverse-engineering pipeline that drives Ghidra in-process and orchestrates Claude/GPT-4o to recover function names, types and structs from stripped, obfuscated binaries. Includes agentic deobfuscation, signature matching, eval harness and cost tracking.
Use Cases
Analyze stripped and obfuscated binaries autonomouslyRecover function names, types and parameter names via LLM-guided decompilationDeobfuscate control-flow flattening, bogus control flow and string encryptionBuild call-graph-ordered context windows for LLM analysisSynthesize struct definitions from field access patternsMatch standard library and crypto function signaturesWrite recovered symbols back into Ghidra's program databaseExport structured analysis.json and decompiled C sourceEvaluate symbol and type accuracy against ground-truth sourceTrack token usage and cost per provider/model
Built With
- Language
- Python
- Frameworks
- Ghidra · PyGhidra · JPype · Anthropic SDK · OpenAI SDK · z3 · Click · Textual · Rich · LIEF · pytest · hatchling · uv
Tags
reverse-engineering · binary-analysis · ghidra · deobfuscation · decompilation · agentic-ai · llm-orchestration · type-recovery · symbol-recovery · symbolic-execution · cli · tui · evals · cost-tracking · pyghidra · call-graph-analysis