Dynatrace Intelligence DQL generation model
- Latest Dynatrace
- Explanation
- 2-min read
- Published May 07, 2026
To ensure that Dynatrace Intelligence provides the most accurate and efficient queries, we use a custom, fine-tuned model for DQL generation. This model is trained on internally sourced DQL queries and is intrinsically knowledgeable of how DQL works. This means that:
- We don't need to augment each user prompt with information about DQL, available data models, or operators to create DQL queries.
- Dynatrace Intelligence generates more efficient and accurate DQL queries.
This custom model is used to:
- Query with natural language in
Dashboards and 
Notebooks. - Generate queries in Dynatrace Assist (if you have enabled Dynatrace agentic AI).
- Generate queries via the MCP server with the help of
Grail Query Agenttool.
Model details
This large language model (LLM) is derived from Meta Llama-3.1-8B. Additional information regarding the terms of use can be found under Llama 3.1 Community License Agreement. Built with Llama.
The underlying foundation model and training data for the DQL generation model may be updated or replaced at any time as we continue to improve Dynatrace Intelligence.
Training data
Dynatrace uses its own internal datasets to train our DQL generation model.
We don't use customer data or any publicly available datasets at any stage of the training process.
You can request additional information regarding training content or data flow diagrams, under a non-disclosure agreement (NDA), via our Trust Center Portal.
The DQL generation model uses AI-generated data for training.