ADK Samples is a collection of ready-to-use agents built on Google’s Agent Development Kit (ADK). My fork focuses on the security-oriented samples — fact-checking LLM output, retrieval-augmented generation, and cloud-security integration — which line up directly with my work as an AI and security specialist at Google Cloud.
The concepts
The repository shows a spread of agent complexity, from a single-purpose auditor to multi-agent workflows. Three samples stand out. The LLM Auditor is an automated fact-checking layer: it identifies factual claims in a model’s response, verifies them with Google Search and internal knowledge, and can rewrite the answer to correct inaccuracies. The RAG agent answers questions about documents uploaded to Vertex AI RAG Engine and returns answers with citations back to the source. The Wiz MCP server exposes the Wiz cloud-security platform as a Model Context Protocol server, so an AI assistant can query cloud-security findings through a standard interface.
How to use it
Each agent lives in its own directory under python/agents/ with a dedicated README covering setup and configuration. You pick the sample you want, install its dependencies with Poetry, supply the relevant Google Cloud and agent-specific credentials, and run it through the ADK CLI or web interface.
Running it
git clone https://github.com/google/adk-samples.git
cd adk-samples/python
You need Python 3.9+, Poetry, and a Google Cloud account for most agents. These samples are for demonstration rather than production use.
Language: Python Source code: github.com/avnit/adk-samples Upstream: github.com/google/adk-samples