AI Potluck
Contribute
Back to Gap Map Product / UX / Safety & Guardrails

Giskard

Giskard
open source / Overall score: 3.0

Giskard is a Python library from Giskard AI for testing and red-teaming LLM and agentic systems, rewritten for dynamic, multi-turn testing of agents. Its vulnerability scan automatically probes an application for prompt injection, harmful output and hallucination, and Giskard Checks runs scenario-based evaluations, including multi-turn agent validation. The scan ships in the giskard-scan package alongside knowledge-base and RAG evaluation.

The repository is now Giskard-AI/giskard-oss, and links to the old Giskard-AI/giskard path redirect there.

Openness

5 high confidence
5.0
license
Apache-2.0(OSI)
source
public
self-host
yes
service
proprietary(Giskard Hub enterprise platform, separate from the library)
core features withheld
no

Giskard's library is published whole under Apache-2.0 and runs locally, using whichever LLM provider the user picks as the judge. Giskard AI also sells an Enterprise platform with its own red-teaming probes and deployment options, but it is a separate product and nothing is withheld from the library itself.

  • https://docs.giskard.ai/oss recorded 2026-09-29

    The library docs describe "a Python package for testing, evaluating, and scanning agentic applications" whose tests "run on your machine, using the LLM provider of your choice as the judge", with a vulnerability scan and a quality scan.

  • https://github.com/Giskard-AI/giskard recorded 2026-08-13

    Redirects to Giskard-AI/giskard-oss, whose metadata records license spdxId - Apache-2.0. The README installs the whole thing with pip and documents running it locally; the only paid-adjacent ask on the page is GitHub sponsorship, and there is no enterprise directory, commercial edition or license key, so nothing is withheld from the published source.

  • https://pypi.org/pypi/giskard/json recorded 2026-09-29

    PyPI metadata for giskard 3.0.0 carries the classifier "License :: OSI Approved :: Apache Software License", lists a Source Code link to the public Giskard-AI repository, and its README calls the project "Giskard is an open-source Python library for" testing agents, installed whole with pip. Searched the body for enterprise, license key and paid: none.

  • https://www.giskard.ai/pricing recorded 2026-09-29

    The pricing page lists a free tier with the "Open-Source library" and "Local deployment", beside a separate Enterprise platform sold with "Hybrid deployment options" and multi-turn red-teaming probes.

Adoption

2 low confidence
2.0

Giskard's adoption reading comes from its GitHub star count, and a star is not a use, so the reading cannot climb from that signal alone. No other usage figure -- downloads, installs, or customer count -- appears in the sources checked here.

Capability

4 medium confidence
4.0

Giskard pairs an automated red-team scan, covering prompt injection, harmful content, stereotypes and misinformation, with a scenario-based evaluation library that tests full multi-turn conversations and RAG grounding.

  • https://docs.giskard.ai/oss recorded 2026-09-29

    The library docs describe "a Python package for testing, evaluating, and scanning agentic applications" whose tests "run on your machine, using the LLM provider of your choice as the judge", with a vulnerability scan and a quality scan.

  • https://github.com/Giskard-AI/giskard recorded 2026-08-13

    README describes 'Evals, Red Teaming and Test Generation for Agentic Systems' and a vulnerability scanner covering 'red teaming, prompt injection, jailbreaks & harmful content' (vulnerability_scan, successor of v2 Scan) plus knowledge-base and RAG quality evaluation, with v3 built for dynamic multi-turn testing of agents.

  • https://pypi.org/pypi/giskard/json recorded 2026-09-29

    The README describes v3 as a "fresh rewrite designed for dynamic, multi-turn testing of AI agents"; Giskard Scan generates adversarial suites "covering prompt injection, harmful content, stereotypes, misinformation, and more", and Giskard Checks lists "Evaluate multi-turn agents" and RAG grounding checks.

Verified 2026-09-27