Chatting with ReVa - An unplanned demo

A quick demo of ReVa the Reverse Engineering Assistant. The program is from a router. I don't really have a goal, just testing with some questions!

[@cyberkaida]

If you would like to see the code, it is available on Github with the Apache 2.0 license! github.com/cyberkaida/reverse-engineering-assistant

[@happimouser1841]

:0 thats soo cooolll :000

[@1AEGIS]

first time hearing of this

[@daa2622]

coooll

[@fenix5fire]

Actually looks interesting

[@cyberkaida]

Thank you! I have been working on this about one day a week for the last 10 months, ReVa is quickly becoming ready for harder tasks! It works well for CTF challenges and malware analysis now!

[@takkeshi_dev]

Looks baller ngl, I'm guessing it's fine-tuned, what is ur base model?

[@cyberkaida]

ありがとうございます！It is not fine tuned, this is deep integration with Ghidra. In this video I am using GPT-4o, but ReVa works with Ollama for offline and private use. I have tested with mixtral, also with good results!

[@takkeshi_dev]

@@cyberkaida niceee, good alternative to binary ninja assistant, keep up the good work

[@RazoBeckett.]

Woh

[@cyberkaida]

Thank you! I am glad you are interested!

[@Piceco-zw8jt]

What are you using to render the chat in the terminal? Just Langchain?

[@cyberkaida]

This is custom! I built it with rich and prompt-toolkit for the rendering. I use gRPC to communicate between the terminal and the Ghidra extension. I inject hooks into langchain to extract the steps the LLM takes in real time and send those to be rendered over gRPC. I am porting to langgraph right now, so the thoughts are not appearing in real time, but I will fix that soon!