giskard.ai

# Giskard AI

> Giskard provides the first automated red teaming platform for conversational AI agents, proactively detecting both security vulnerabilities and business compliance failures, from prompt injection and data disclosure to hallucinations and inappropriate denials.

Important notes:

- Giskard provides a LLM vulnerability scanner to secure conversational AI agents that conduct dynamic, multi-turn attacks across dozens of vulnerability categories covering more than 40 probes.
- Giskard Hub UI is a platform for business users to create test datasets, run evaluations, and team collaboration.
- Giskard also provides a Hub SDK for developers with a Python interface to interact with the hub programatically. It allows to automate testing workflows, integrate in CI/CD pipelines, and build custom tools.
- It integrates with Hugging Face, Databricks, Nvidia, OpenAI, Mistral AI, GitHub, etc.
- Giskard allows flexible installation on premise or cloud (AWS) environment.

## Docs

- [Giskard documentation]
(https://docs.giskard.ai/): Official documentation with installation, tutorials, and API reference.
- [LLM vulnerability scanner](https://docs.giskard.ai/hub/ui/scan/index.html): A quick start of how to run the scan to detect AI vulnerabilities.
- [Giskard quick start](https://docs.giskard.ai/hub/sdk/index.html): A brief overview of how to set up the Giskard Hub SDK.

## Additional resources

- [Giskard GitHub]
(https://github.com/Giskard-AI/giskard-oss): Open source library code, issues, and community contributions.
- [Phare](https://phare.giskard.ai/): a multilingual benchmark to evaluate LLMs across key safety & security dimensions, including hallucination, factual accuracy, bias, and potential harm.
- [Giskard’s RealHarm]
(https://realharm.giskard.ai/): a dataset of problematic interactions with textual AI agents built from a systematic review of publicly reported incidents.
- [Giskard’s RealPerformance]
(https://realperformance.giskard.ai/): a dataset of functional issues of language models, that mirrors failure patterns identified through rigorous testing in real LLM agents.