Intent Recognition

The Intent Recognition node enables AI agents to identify users' intent from their input (Query), and route different intents to distinct branches within the workflow. This enhances the user experience and improves the agent's practical effectiveness in real-world applications.

One user input will only be recognized as one intent.

Node Configuration

• LLM: The large language model used for intent recognition.

• Query: The input text to be analyzed for intent recognition. The system will identify the intent of the Query (leveraging the LLM's capabilities) and match it to the most relevant intent.

• Short-term Memory: If the current workflow is embedded within an agent, the agent’s dialogue history (i.e., memory) can be included as part of the context for intent recognition.

• Intent Options: Define multiple distinct intents in natural language as the target for Query recognition.

Node Output

The Intent Recognition node outputs the following two types of parameters after execution:

• classificationId: The ID of the matched intent. IDs are assigned sequentially based on the order of configured intents (the first intent has an ID of 1). If no configured intent is matched, the ID defaults to 0, triggering the "Other" branch.

• reason: The reason and basis for the classification, automatically generated by the LLM.