How to create a diagnostic agent using Gemini and osquery

Talk presented at Biznagafest (DevFest Malaga) in October 2025

Inspired by the "computer" in Star Trek, in this talk we are going to create a natural language diagnostic agent that can answer questions about your computer, from asking the number of cores and amount of memory installed to running a "Level 1 Diagnostic Procedure". This AI agent will be built step by step using Python, Vertex AI, Gemini and OSQuery. OSQuery is an open source tool originally developed by Facebook to expose operating system information using the SQL language. Its benefits include the ability to perform rapid diagnosis using queries instead of memorising command line tools and joining data manually. It's also possible to download query packs for the most common diagnostic procedures. As LLMs are really good at writing SQL, we are going to explore how this tool in the "hands" of an AI agent can enable advanced diagnostics using natural language, effectively mimicking the USS Enterprise engineering experience.

Transcript

1. How to create a diagnostic agent using Gemini and osquery Daniela Petruzalek Developer Relations Engineer @ Google Cloud @danicat83
2. Who am I? Developer Relations Engineer at Google UK Originally from Brazil 󰎙, but living in the UK 󰏅 since 2019. My background is in backend and data engineering.
3. Agenda Motivation The technology Diagnostic agent in action
4. How will the engineering of the future look like?
5. https://www.tomshardware.com/tech-industry/artificial-intelligence/jensen-huang-advises-against-learning -to-code-leave-it-up-to-ai Published 25/Feb/2024 https://www.finalroundai.com/blog/sam-altman-says-world-wants-1000x-more-software Published 24/Jul/2025
6. Space: The Final Frontier… “Hello computer” “Computer, execute a level 1 diagnostic procedure” “Computer, run a simulation” “Computer, create a routine”
7. What if we could do this today?
8. “Computer, run a level 1 diagnostic procedure” Engineer Agent Agent ??? Report
9. OSQuery osquery is a monitoring, instrumentation and analysis tool for operating systems using the SQL language Available for Linux, macOS and Windows
10. Agent Development Kit Agent Development Kit (ADK) is a framework for development and deployment of AI agents. While it is optimised for Gemini and Google products, ADK is model-agnostic and environment-agnostic https://google.github.io/adk-docs/
11. Agent Categories
12. Diagnostic Agent V1
13. Tool Definition
14. Testing the Agent Command line: adk run [agent] Dev-UI: adk web API server: adk api_server
15. Problems! osquery tables vary according to the host operating system Agent may produce “valid” but incorrect queries: SELECT * FROM process WHERE name = "malware" Hallucinations: models can generate fake information
16. Context Engineering The model’s response can be only as good as the context it holds Improvement points: - System prompt - Tools (!!!) - Information retrieval: - Web Search - Retrieval Augmented Generation (RAG)
18. V2: An Improved System Prompt
19. V3: Connect With the Outside World
20. V4: Schema Discovery Ideia: supply the list of tables and respective schemas using RAG Schemas: github.com/osquery/osquery/tree/master/specs Vertex AI RAG Engine: cloud.google.com/vertex-ai/generative-ai/docs/rag-engine/rag-overview
21. Bonus: It’s aliiiive! Gemini Live API (experimental): gemini-live-2.5-flash-preview-native-audio cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-live
22. V5: Tidying it up Refactor with SequentialAgent Three separate agents: Query Planner -> Executor -> Analyser
23. What about the client?
24. github.com/sito-sikino/avatar-ui-core
25. Final Thoughts Developing agents can be challenging, but tools like ADK can ease some of the pain Developing clients as a backend engineer is still hard, but coding agents can help (e.g. Gemini CLI) Diagnosing a computer just with voice is not a reality yet, but it seems pretty close
26. Useful Resources Code: github.com/danicat/biznagafest ADK: google.github.io/adk-docs ADK samples: github.com/google/adk-samples Models: ai.google.dev/gemini-api/docs/models Osquery: www.osquery.io Vertex AI RAG: cloud.google.com/vertex-ai/generative-ai/docs/rag-engine/rag-quickstart
27. Thank you! danicat.dev linkedin.com/in/petruzalek @danicat83