Dynatrace offers an AI Observability product that lets enterprise customers monitor the performance, cost, and safety of their own AI-powered applications — including tracking how large language models (LLMs) are behaving, what they cost to run, and whether they are producing harmful or non-compliant outputs.

Announced in January 2025 and expanded since, the AI Observability platform supports over 20 AI and LLM technologies including OpenAI, Anthropic, Amazon Bedrock, and Google Gemini. It collects metrics on token usage, response latency, error rates, and guardrail activations (e.g., PII leakage, hallucinations, prompt injections, toxic language). Davis AI's predictive capabilities analyze usage behavior to forecast cost changes associated with LLM usage. The platform also provides multi-model tracing — mapping dependencies between multiple LLMs in Retrieval Augmented Generation (RAG) pipelines or agentic AI frameworks — and maintains a full audit trail of every input and output to support regulatory compliance such as the EU AI Act.