Skip to main content
In Cognigy, intents identify users’ goals based on their input. Each intent is defined by example sentences that show how users might express a specific request. For example, an intent named OrderFood might include sentences like Can you get me some food? or I'd like to order a pizza. The NLU engine analyzes and scores these intents to match user inputs. Each intent includes:

Work with intents

You can create, edit, and manage intents, as well as download and upload them, in Advanced > NLU on the Intents tab.

Download and upload intents

You can upload intents to and download them from Cognigy as CSV or JSON files. Use these files to import intents, edit them using your favorite spreadsheet or developer tools, and move them between flows.
Your CSV should be formatted as follows:
  • Don’t include a header row.
  • Cognigy automatically detects various CSV formats. If you encounter any issues, use UTF-8 encoding, a comma (,) as the delimiter, double quotes (") as your quote character, and newline (\n) as your line terminator.
  • You’re required to fill in the following columns:
CSV format example:
Keep regional settings consistent on your operating system (OS), as they affect the CSV delimiter. Mismatched settings can break the file and cause upload failure.
The JSON format is a comprehensive specification for defining intents in Cognigy.The following JSON example illustrates the structure and content of an intent, including the intent name (Pizza), example sentences, rules, tags, disambiguation sentence, default reply, and other intent-related settings.JSON format example:

Add example sentences

When you create an intent, add example sentences that help the NLU model recognize user queries.

Build the NLU model

Whenever you make changes to the intent or its related settings, you need to rebuild the NLU model. If the NLU model needs rebuilding, the red indicator appears:
  • Next to the flow name in the flows list
  • Next to the Build Model button in Advanced > NLU
If your flow includes states and intents with intent conditions, you can exclude them from training using Quick Build. This approach speeds up the NLU model building process, especially for large flows that contain many states and intent conditions. If you don’t use these features, both Build Model and Quick Build deliver the same model quality and training speed. You can build the NLU model for a single flow or for all flows in your project:
  1. In the navigation menu, go to Advanced > NLU.
  2. In the flow selector at the top, select the flow that you want to train.
  3. On the Intents tab, click Build Model or click the dropdown arrow next to it and select Quick Build.
If you’ve attached flows to the current flow, the model is automatically built for the attached flows. In the UI, trained attached flows appear as untrained (with a red indicator), which you can ignore.
  1. In the navigation menu, go to Design > Flows.
  2. On the Flows page, click Train all Flows. If Train all Flows isn’t available, for version 2026.2 or earlier, set FEATURE_TRAIN_ALL_PROJECT_FLOWS: "true" in values.yaml for on-premises installations, or upgrade to version 2026.3 for SaaS.
Note that during training, flows are retrieved from the database sequentially, following their default storage order. Training doesn’t run in parallel. Each flow is trained one after another, regardless of the number of service replicas or deployed locales.If the NLU model is built successfully, you receive a green success notification. Otherwise, an error notification appears with details. For example, you may need to add more example sentences.

Apply additional settings

More information

Last modified on November 11, 2026