https://github.com/sha0coder/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%
https://github.com/sha0coder/neurostr
Last synced: 4 months ago
JSON representation
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%
- Host: GitHub
- URL: https://github.com/sha0coder/neurostr
- Owner: sha0coder
- Created: 2022-05-26T17:47:39.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2022-05-26T18:06:19.000Z (over 3 years ago)
- Last Synced: 2025-03-31T07:11:10.365Z (9 months ago)
- Language: Rust
- Size: 935 KB
- Stars: 1
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# 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)
```bash
~/s/neurostr ❯❯❯ target/release/neurostr exec model.ai /bin/ls | wc -l main ✱
510
~/s/neurostr ❯❯❯ strings /bin/ls | wc -l main ✱
1608
```
## Usage
```bash
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.
```bash
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.