Some products and features are in the process of being renamed. Generative playbook and flow features are also being migrated to a single consolidated console. See the details.
Stay organized with collections
Save and categorize content based on your preferences.
During a conversation,
Conversational Agents (Dialogflow CX) agents always use language models for understanding end-user intention,
but you can choose whether and how language models are used for agent responses.
You can decide between fully generative, partly generative,
and deterministic features when designing your agent.
This guide provides an overview of these features.
It helps to decide which of these features
you plan to use,
so you know which documentation will be relevant to you.
Fully generative
The fully generative features are built on Vertex AI
large language models (LLMs) for both understanding end-user intention
as well as generating agent responses.
These features are easy to use and provide a very natural conversation.
The following is an overview of the fully generative features:
X
Item
Playbooks
Playbooks provide a new way for creating virtual agents using LLMs. You only need to provide natural language instructions and structured data. This can significantly reduce the virtual agent creation and maintenance time, and enable brand new types of conversational experiences for your business.
Data stores
Data stores parse and comprehend your public or private content (website, internal documents, and so on). Once this information is indexed, your agent can answer questions and have conversations about the content. You just need to provide the content.
Deterministic flows
If you require more deterministic control over the conversation
and all responses generated by the agent,
you can design your agent with flows.
X
Item
Flows
Flows use language models for understanding end-user intention during a conversation, which may not be completely deterministic. However, once intention is established, you have complete control over the conversation flow and agent responses. Designing an agent with deterministic flows typically takes more design time, but this is a good option for agents that require explicit control over agent responses.
Partly generative flows
Flows have some optional generative features that you can use when you don't
need deterministic control over agent responses in certain conversation scenarios.
The following is an overview of these features:
X
Item
Generators
Generators are used to generate agent responses. Rather than providing the agent response explicitly, you provide a LLM prompt that can handle many scenarios including conversation summarization, question answering, customer information retrieval, and escalation to a human.
Generative fallback
Generative fallback is used to generate agent responses when end-user input does not match an expected intention. You can enable generative fallback in certain scenarios by providing a LLM prompt to generate the response.
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Hard to understand","hardToUnderstand","thumb-down"],["Incorrect information or sample code","incorrectInformationOrSampleCode","thumb-down"],["Missing the information/samples I need","missingTheInformationSamplesINeed","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2025-08-07 UTC."],[[["\u003cp\u003eConversational Agents use language models to understand user intentions, but agents can be designed to be fully generative, partly generative, or deterministic in how they respond.\u003c/p\u003e\n"],["\u003cp\u003eFully generative features use large language models (LLMs) for both understanding user intent and generating agent responses, providing a natural conversational experience through features like Playbooks and Data Stores.\u003c/p\u003e\n"],["\u003cp\u003eDeterministic flows offer complete control over the conversation and agent responses, using language models for understanding intent but giving you explicit control once intent is established.\u003c/p\u003e\n"],["\u003cp\u003ePartly generative flows allow for optional generative features like Generators and Generative Fallback, leveraging LLMs to handle various scenarios, such as summarization or generating responses when user input doesn't match expected intentions.\u003c/p\u003e\n"],["\u003cp\u003eChoosing between fully generative, partly generative, and deterministic features depends on the level of control needed over agent responses and the desired conversational experience.\u003c/p\u003e\n"]]],[],null,["# Generative versus deterministic\n\nDuring a conversation,\nConversational Agents (Dialogflow CX) agents always use language models for understanding end-user intention,\nbut you can choose whether and how language models are used for agent responses.\nYou can decide between fully generative, partly generative,\nand deterministic features when designing your agent.\n\nThis guide provides an overview of these features.\nIt helps to decide which of these features\nyou plan to use,\nso you know which documentation will be relevant to you.\n\nFully generative\n----------------\n\nThe fully generative features are built on [Vertex AI](/vertex-ai/docs)\nlarge language models (LLMs) for both understanding end-user intention\nas well as generating agent responses.\nThese features are easy to use and provide a very natural conversation.\nThe following is an overview of the fully generative features:\n\nDeterministic flows\n-------------------\n\nIf you require more deterministic control over the conversation\nand all responses generated by the agent,\nyou can design your agent with flows.\n\nPartly generative flows\n-----------------------\n\nFlows have some optional generative features that you can use when you don't\nneed deterministic control over agent responses in certain conversation scenarios.\nThe following is an overview of these features:"]]
