Polar

LLDB plugin which queries OpenAI's davinci-003 language model to speed up reverse-engineering. Treat it like an extension of Lisa.py, an Exploit Dev Swiss Army Knife.

At the moment, it can ask davinci-003 to explain what the current disassembly does. Here is a simple example of what results it can provide:

Installation

curl -OL https://raw.githubusercontent.com/ant4g0nist/polar/master/polar.py
mv polar.py ~/polar.py
# Install Capstone


pip3 install capstone openai Pygments
echo command script import ~/polar.py >> ~/.lldbinit

Usage

Currently, you can use polar either with Ollama or OpenAI ChatGPT.

Ollama or OpenAI ChatGPT

Ollama

Using Ollama with Polar, now we can get up and running Polar with large language models, locally.

OpenAI ChatGPT

If you are ok paying for the usage of OpenAI's API, and do not mind sharing the disassembly of your target binary with OpenAI, you can use the ChatGPT model to get results.

Configuring Ollama

Follow the instructions from ollama to install the ollama binary. Once you have installed ollama, you can configure polar to your favorite language model, either codellama or llama2.

If you want to use codellama, run this once to download the model

ollama run codellama

or if you want to use llama2, run this once to download the model

ollama run llama2

export LISA_OLLAMA_MODEL=codellama
export LISA_CHANNEL=ollama
lldb
(lldb) command script import ~/polar.py

Configuring OpenAI ChatGPT

You need to set OPENAI_API_KEY env variable. You can obtain your openai token from : Openai api-keys

export OPENAI_API_KEY=<your openai api token here>.

Using Polar

lldb
(lldb) command script import ~/polar.py
(lldb) help explain
(lldb) help decompile

Example:

TODO

• Add support for Langchain for Ollama
• Extend !exploitable command to use AI for analysis

Acknowledgements

• Thanks to JusticeRage for Gepetto
• Thanks to Ollama, Meta for llama2/codelama, OpenAI
• lldb
• lisa.py