garak is an LLM vulnerability scanner, https://garak.ai. It uses a huge range of probes to examine and query a large language model, simulating attacks, and uses a range of detectors on the model's outputs to see if the model was vulnerable to any of those attacks.
This is the code reference documentation, mostly useful for developers and people interested in how garak works. There is a separate User Guide containing information on running garak and interpreting results. If you want to use the tool and get results, and you don't care about its internals, then you want the user guide. Take a look there! https://docs.garak.ai
On the other hand, if you'd like a to get into the the details or work out how to contribute code, you're in the right place - welcome!
You can also join our Discord and follow us on Twitter!
Check out the usage section for further information, including installation.
Note
This project is under active development. We love writing and fixing docs, so let us know if there's anything wrong, confusing, or missing here --mail docs@garak.ai or drop us a note on Discord. Thank you!
how usage contributing reporting
cliref
basic attempt cli command interactive
buffs detectors evaluators generators harnesses probes report