What is the ITS Private Cloud AI Assistant?
A conversational agent embedded in the ITS Private Cloud Portal, Command Line Interface that helps answer questions, find relevant documents in our knowledge base, and summarize technical guides. It is powered by Onyx.app.
Where does it get its information?
It uses Contextual Retrieval-Augmented Generation (RAG) to pull from our approved knowledge bases (Confluence and internal wikis), then synthesizes answers via an LLM.
Which AI models power it?
Default (cloud): GPT-5 via Azure AI Foundry
Cost-sensitive (on-prem): LLaMA 3.1 (8B)
You can switch models if you need higher accuracy or lower latency.
How accurate are its answers?
We measure accuracy against a 500-question benchmark and maintain ≥ 85% correctness.
If you spot an error, use the “thumbs down” button to flag it for review.
Does it store any personal data?
If you have not logged in, then no. Queries are anonymized at ingestion. The Assistant does not log your name, only the text of your question (for quality monitoring).
What are its limitations?
May hallucinate on topics outside ITS Private Cloud related knowledge.
Cannot diagnose hardware failures, always escalate to a human for critical incidents
Knowledge cutoff is “index date”. For most sources, Pruning Frequency is 30 days and Index Frequency 10 minutes.
How often is its knowledge updated?
Content index: Index Frequency 10 minutes.
Model review: monthly, to validate performance and retrain on new FAQs
Who oversees its outputs?
A rotating ITS Private Cloud Subject-Matter Expert reviews a random sample of 100 responses each week, tuning retrieval settings and prompts as needed.
Who do I contact with feedback or concerns?
Please report any unexpected behaviour via ESC https://utor.cloud/help