# Giskard

> Automated testing platform for LLM agents that detects hallucinations, security vulnerabilities, and quality issues through continuous red teaming.

Giskard provides an automated testing platform designed to secure LLM agents by detecting hallucinations, security vulnerabilities, and quality issues before they reach production. The platform combines continuous red teaming with expert guidance to help enterprise teams find and fix security gaps in their AI applications. Trusted by major companies including L'Oréal, AXA, Société Générale, Decathlon, Michelin, and BNP Paribas, Giskard serves as a safety net for organizations deploying conversational AI agents.

- **Continuous Red Teaming** automatically generates sophisticated attack scenarios to uncover vulnerabilities that manual audits miss, delivering the largest test coverage of both security and quality issues with high domain specificity.

- **Hallucination Detection** leverages internal knowledge sources like RAG knowledge bases to identify factual errors, contradictions, omissions, and unsupported information in AI responses.

- **Security Vulnerability Testing** detects critical issues including prompt injection attacks, data disclosure risks, sycophancy attacks, and inappropriate content generation using security vulnerability taxonomies.

- **Quality Assurance** catches business failures like inappropriate denials where bots incorrectly refuse valid questions, ensuring AI agents deliver helpful responses without being overly restrictive.

- **Human-in-the-Loop Dashboards** enable business, engineering, and security teams to review, customize, and approve tests through a collaborative visual interface with annotation tools.

- **Regression Prevention** transforms discovered vulnerabilities into permanent protection by automatically converting detected issues into reproducible test suites that enrich golden test datasets.

- **Python SDK & Web UI** allows teams to execute tests programmatically or schedule them through the web interface to ensure AI agents meet requirements after each update.

- **Sovereign & Secure Infrastructure** offers data residency choices (EU or US), role-based access control, audit trails, identity provider integration, end-to-end encryption, and compliance with GDPR, SOC 2 Type II, and HIPAA.

To get started, teams can request a trial through the Giskard website. The platform operates as a black-box testing tool, meaning it only requires API endpoint access to the AI agent without needing knowledge of internal components like foundation models or vector databases.

## Features
- Continuous Red Teaming
- Hallucination Detection
- Prompt Injection Detection
- Data Disclosure Prevention
- Sycophancy Attack Detection
- Quality Assurance Testing
- Human-in-the-Loop Dashboards
- Regression Prevention
- Python SDK
- Web UI Scheduling
- Role-Based Access Control
- Audit Trails
- Identity Provider Integration
- End-to-End Encryption
- GDPR Compliance
- SOC 2 Type II Compliance
- HIPAA Compliance
- Data Residency Options

## Integrations
Hugging Face, AWS, Google Cloud, Microsoft Azure, Mistral AI

## Platforms
WEB, API, DEVELOPER_SDK

## Pricing
Open Source, Free tier available

## Links
- Website: https://www.giskard.ai
- Documentation: https://docs.giskard.ai/
- Repository: https://github.com/Giskard-AI/giskard
- EveryDev.ai: https://www.everydev.ai/tools/giskard