Davis CoPilot™, Dynatrace's generative AI, is designed to boost productivity, help with onboarding, and enable you to explore data through natural language.
Davis CoPilot is based on a large language model (LLM). The model used by Davis CoPilot generates responses based on your inputs and is probabilistic. This means that the responses are generated by predicting the most probable next word or text, based on the data that they have been created with, and on the provided context. Davis CoPilot uses the Retrieval Augmented Generation (RAG) approach to provide a foundational LLM with the proper context to transform natural language into a DQL query (in-context learning).
Because of this approach, these models can sometimes behave inaccurately, incompletely, or unreliably. This means there is a risk that the response you receive does not accurately reflect the prompt you provided or that the generated content sounds reasonable but is incomplete or inaccurate.
We recommend that you carefully evaluate the responses you get back from Davis CoPilot. If Davis CoPilot responds inaccurately, please provide feedback directly from Notebooks , Dashboards
, or Davis CoPilot Chat .
The Davis CoPilot service offers distinct and specialized skills. Currently, Davis CoPilot offers two skills:
This is why Quick Analysis in Notebooks and Dashboards
cannot answer general questions, and why Davis CoPilot Chat chat might produce inaccurate DQL queries.
Davis CoPilot uses the Retrieval Augmented Generation (RAG) approach to provide a foundational LLM with the proper context to transform natural language into a DQL query (in-context learning). This means that Davis CoPilot will enrich your prompt with relevant additional content or context that is sent to the foundational LLM in order to generate an appropriate response. The content or context that is used to enrich your prompt depends on which underlying skill is queried.
The data and additional context is used only to enrich prompts; the model does not learn from this. Customer data isn't used to automatically fine-tune, train, or improve any models or services. For more information, see How are NL2DQL responses generated?
There is no limit to the number of interactions you can have with Davis CoPilot. However, there is a throughput limit. This means that you can ask 10 questions in a rolling 15-minute timeframe.
There is a similar limit to how many questions can be asked by all users in your environment simultaneously. Your environment can handle up to 40 questions in a rolling 15-minute timeframe.
If you have exceeded any of the limits, you'll see an error message: "I'm sorry, but I couldn't generate an answer for you because of unusually high demand. Please try again in a minute."
If you would like to learn more about Davis CoPilot, visit our FAQ page.