Giskard

Visit Website

Introduction

Giskard is a French company building testing and security tools for AI agents. Its open-source library, now at Giskard-AI/giskard-oss under Apache-2.0, describes itself as “Evals, Red Teaming and Test Generation for Agentic Systems”. Giskard v3 is a rewrite for dynamic, multi-turn testing of agents; v2 remains available but is no longer actively maintained.

The commercial platform covers agent red-teaming, agent evaluation and agent guardrails, with customer stories such as a bank securing its customer-facing chatbot. The company states GDPR adherence, SOC 2 Type II and HIPAA compliance.

Features

1. Checks library: pip install giskard installs checks for agents and LLM apps.
2. Vulnerability scan: giskard[scan] adds an AI vulnerability and quality scan, including RAG evaluation.
3. Multi-turn testing: v3 is built for dynamic, multi-turn agent tests.
4. Red teaming: Commercial agentic red teaming with regularly updated adversarial techniques and OWASP-style coverage.
5. Domain datasets: Generation of domain-specific evaluation datasets.
6. Human review: Collaborative interfaces for human-in-the-loop review.
7. Guardrails: Continuous protection of agents in production (commercial).

Pros

1. Open-source core: Apache-2.0 library.
2. Security plus quality: Red teaming and evals in one vendor.
3. European vendor: GDPR-native for EU buyers.
4. Lightweight v3: Fewer heavy dependencies than v2.
5. Research output: Whitepapers and published comparisons.

Cons

1. Python 3.12+: v3 requires a recent Python.
2. v2 to v3 migration: Older tabular scans remain v2-only.
3. Commercial features gated: Hub features need a paid plan; pricing is not listed.
4. LLM judge costs: Some checks use model calls.
5. Fast change: APIs are still evolving.

Usage and Links (Example)

From the README:

pip install giskard
pip install "giskard[scan]"
pip install "giskard[openai]"

Write checks for your agent’s expected behaviour, run the scan against your LLM app, and add the suite to CI. See docs.giskard.ai for examples.

Resources:

• Giskard website
• GitHub repository
• Docs

Pricing

The open-source library is free under Apache-2.0. The commercial Hub is sold through demos; the pricing page lists features rather than prices. See site.

Recommended for?

1. Teams shipping LLM agents that need both quality evals and red teaming.
2. EU organisations that prefer a European vendor.
3. Python developers wanting an open-source checks library.
4. Not a fit if you need a no-code tool with public pricing.

Conclusion

Giskard combines an open-source testing library with a commercial red-teaming and guardrails platform for AI agents.

Start with the library and move to the Hub when you need collaboration and continuous protection.