/neurostr

linux strings command on memory blobs generates many crap, this tool uses a neural network and a trained model for reducing the crap in 85%

Primary LanguageRust

NeuroStr neural network based strings

Description

Doing strings command to a memory dump or binary file generates many amount of crap strings. This tool use a trained neural network, and shows the strings, also show some crap but reduces the crap in 85% I haven't identified false negatives in several tests, but thre are some false positives (crap strings)

~/s/neurostr ❯❯❯ target/release/neurostr exec model.ai /bin/ls | wc -l                                  main ✱
510
~/s/neurostr ❯❯❯ strings /bin/ls | wc -l                                                                main ✱
1608

Usage

make
target/release/neural exec model.ai memory.dump

Train a model

you need a good.txt with all type of good information, and crap.bin with some crap, it aslso uses random generator to generate random datasets.

make
target/release/neural train good.txt mymodel.ai

is necesary a file named crap.bin but can be empty, this can be used to reduce the false positives you find.

The model model.ai

on the good.txt i used:

  • password wordlist
  • urls
  • domains
  • csv's
  • ip's
  • other information

TODO:

  • memory optimizations.