Workflow node types

With Lorikeet, you compose workflows by linking together different workflow nodes. Each node has a different purpose, optimized to get the most out of the available AI models.

This guide contains an overview of each node and some tips for configuration.

​

Gather

Gather nodes allow you to empower the AI to get information by talking to users, doing analysis, or calling APIs.

​

Basics

Gather nodes aim to ‘gather’ the data you configure it to get. When the node has gathered the data it will pass on to the next node in the workflow. If it cannot gather the data after multiple attempts, it will escalate.

You first define each piece of data you want the AI to gather.

Then you can provide instructions for how it should go about doing this. It can gather data through conversation with the customer, or by calling APIs you provide.

​

Configuration tips

• If the node cannot find a value for the data you tell it to gather, it will escalate. So if you believe data may not be availble, but you want the AI to proceed, tell it how to deal with missing data. For instance, if an API return may be null, tell the AI to populate the field with ‘not available’ if you want the AI to proceed.
• Aim to configure data to gather that will only be populated when the AI has completed the actions you asked it to take. For instance if you ask it to gather ‘has action been taken’ it could set that to false without taking any action. Instead configure ‘which action was taken’, so it will be populated after the action is taken.
• When asking the AI to call APIs, give it instructions about how to should do so. For instance, you can prompt it about the types of inputs the API expects, or how it should reformat inputs it gets from the user.

​

Analyze

Analyze is the sister node to Gather but rather than chatting to the customer it works in the background to infer or transform data.

​

Basics

Like Gather, Analyze will derive new data from the existing conversation and make it available to nodes further down the workflow.

You define and describe each of the types of data you want it to create and the AI will do the rest.

​

Configuration tips

• You can ensure richer data by augmented the type of data your Analyze node is producing. Common types including categories, true/false, dates and numbers.

​

Choice

A Choice node is a decision-making node that allows your workflow to branch based on specific criteria. This node evaluates the data gathered or analyzed in previous steps and directs the workflow down different paths depending on the conditions you set.

An important distinction between Choice and other nodes is that Choice only makes deterministic evaluations - it doesn’t lean on the AI at all. This is useful when the specifics of a scenario are critical - a user must be a specific type of account, a date must have passed, an address needs to be correct.

​

Basics

The Choice node functions by evaluating a set of conditions or rules that you define. Based on these conditions, the workflow can continue down one of several predefined paths. This is particularly useful for creating dynamic workflows that need to adapt to varying inputs or outcomes.

You can define simple conditions, such as checking if a value is true or false, or more complex logic that considers multiple factors.

​

Configuration tips

• When configuring a Choice node, make sure to thoroughly define each condition to avoid unintended behavior.
• Consider the sequence of conditions carefully; the workflow will follow the first matching condition it encounters.
• Use Choice nodes to handle scenarios where the workflow needs to adapt to different user inputs, data states, or external factors.

​

Outcome

An Outcome node is designed to close out a workflow. All workflows will have at least one outcome and probably will have many. By applying power ups to the Outcome nodes, you can configure actions such as replying to a customer, posting an internal note and closing a ticket. These power ups will depend heavily on your workflows logic and the specifics are customizable by you.

​

Configuration tips

• Outcomes can have many closing actions associated with them - they should in most cases have at least one action