reait

RevEng.AI Toolkit

Analyse compiled executable binaries using the RevEng.AI API. This tool allows you to search for similar components across different compiled executable programs, identify known vulnerabilities in stripped executables, and generate "YARA++" REAI signatures for entire binary files. More details about the API can be found at docs.reveng.ai.

NB: We are in Alpha. We support GNU/Linux ELF and Windows PE executables for x86_64, and focus our support for x86_64 Linux ELF executables.

Installation

Install the latest stable version using pip3.

pip3 install reait

Latest development version

pip3 install -e .

or

python3 -m build .
pip3 install -U dist/reait-*.whl

Using reait

Analysing binaries

To submit a binary for analysis, run reait with the -a flag:

reait -b /usr/bin/true -a

This uploads the binary specified by -b to RevEng.AI servers for analysis. Depending on the size of the binary, it may take several hours. You may check an analysis jobs progress with the -l flag e.g. reait -b /usr/bin/true -l.

Extract symbol embeddings

Symbol embeddings are numerical vector representations of each component that capture their semantic understanding. Similar functions should be similar to each other in our embedded vector space. They can be thought of as advanced AI-based IDA FLIRT signatures or Radare2 Zignatures. Once an analysis is complete, you may access RevEng.AI's BinNet embeddings for all symbols extracted with the -x flag.

reait -b /usr/bin/true -x > embeddings.json

Search for similar symbols using an embedding

To query our database of similar symbols based on an embedding, use -n to search using Approximate Nearest Neighbours. The --nns allows you to specify the number of results returned. A list of symbols with their names, distance (similarity), RevEng.AI collection set, source code filename, source code line number, and file creation timestamp is returned.

reait --embedding embedding.json -n

The following command searches for the top 10 most similar symbols found in md5sum.gcc.og.dynamic to the symbol starting at 0x33E6 in md5sum.clang.og.dynamic. You may need to pass --image-base to ensure virtual addresses are mapped correctly.

reait -b md5sum.gcc.og.dynamic -n --start-vaddr 0x33E6 --found-in md5sum.gcc.o2.dynamic --nns 10 --base-address 0x100000

Search NN by symbol name.

reait -b md5sum.gcc.og.dynamic -n --symbol md5_buffer --found-in md5sum.gcc.o2.dynamic --nns 5

NB: A smaller distance indicates a higher degree of similarity.

Specific Search

To search for the most similar symbols found in a specific binary, use the --found-in option with a path to the executable to search from.

reait -n --embedding /tmp/sha256_init.json --found-in ~/malware.exe --nns 5

This downloads embeddings from malware.exe and computes the cosine similarity between all symbols and sha256_init.json. The returned results lists the most similar symbol locations by cosine similarity score (1.0 most similar, -1.0 dissimilar).

The --from-file option may also be used to limit the search to a custom file containing a JSON list of embeddings.

Limited Search

To search for most similar symbols from a set of RevEng.AI collections, use the --collections options with a RegEx to match collection names. For example:

reait -n --embedding my_func.json --collections "(libc.*|lib.*crypt.*)"

RevEng.AI collections are sets of pre-analysed executable objects. To create custom collection sets e.g., malware collections, please create a RevEng.AI account.

Unstripping binaries

Find common components between binaries, RevEng.AI collections, or global search, by using -M, --match.

Example usage:

reait -M -b 05ff897f430fec0ac17f14c89181c76961993506e5875f2987e9ead13bec58c2.exe --from-file 755a4b2ec15da6bb01248b2dfbad206c340ba937eae9c35f04f6cedfe5e99d63.embeddings.json --confidence high

RevEng.AI embedding models

To use specific RevEng.AI AI models, or for training custom models, use -m to specify the model. The default option is to use the latest development model. Available models are binnet-0.1 and dexter.

reait -b /usr/bin/true -m dexter -a

Software Composition Analysis

To identify known open source software components embedded inside a binary, use the -C flag.

Binary ANN Search

To perform binary ANN search, pass in -n and -s flag at the same time. For example:

reait -b /usr/bin/true -s -n
Found /usr/bin/true:elf-x86_64
[
  {
    "distance": 0.0,
    "sha_256_hash": "1d20d8b1bbc861a2e9e0216efb7945fba664a5e6ba5f6a93febd6612a92551a8"
  },
  {
    "distance": 0.04410748228394201,
    "sha_256_hash": "265cb456cf5a09ad82380cb98118fb9255a9c9407085677d597abd828a5f4b11"
  },
  {
    "distance": 0.04710724400903421,
    "sha_256_hash": "1de9c70e46b17a96ee15e88e52da260de4f2d70e167c5172c29416d16f907482"
  },
  {
    "distance": 0.047961843853272956,
    "sha_256_hash": "01bf5e0f03dfaf6324f7e00942fed88ca52845c190a7392b0d0eb5c3a91091df"
  },
  {
    "distance": 0.05086539098571474,
    "sha_256_hash": "62dd31307316ee0e910eb845f35bf548b7fd79dc9f407ef917efdf14d143842e"
  }
]

Configuration

reait reads the config file stored at ~/.reait.toml. An example config file looks like:

apikey = "l1br3"
host = "https://api.reveng.ai"
model = "binnet-0.3-x86"

Contact

Connect with us by filling out the contact form at RevEng.AI.