Voiceflow is an NLU-agnostic agent design and development platform. Because of that, you can design any of the core data types required to build a functional agent and dialogue manager directly with the tools available on your design campus.
If you are designing your agents in Voiceflow, and then handing off these designs off to be implemented on a third-party NLU platform, this process can be painful:
- It requires manual context transfer from designers to developers
- Time is wasted manually re-creating data from Voiceflow in another platform
- Iteration is hard because each change requires a new handoff
To address these issues, we recommend that Conversational AI Teams on Voiceflow build an Exporter to adapt all relevant agent data created as part of their Voiceflow design process into production-ready data sets.
You can categorize a Voiceflow agent into three primary data sets:
- NLU Model
- Intents and Training Phrases
- Entity Types, Values and Synonyms
- Intent-based Routes
- Condition-based Routes
- Event Handlers
- Response Content
- Agent Responses
- Webhook Fulfillment
There is not necessarily 1:1 parity between the functionality you can design into an agent in Voiceflow and your NLU. In this guide, we can help identify how to best include the relevant data sets you’re looking to export for production in your designs using our existing feature sets, and work-arounds where required.
When you have a Voiceflow agent design ready for export to an NLU, the workflow to get your data exported is:
Your Voiceflow agent data structure is represented in a structured JSON format, that includes the individual data sets, and the relationship between the nodes in your design. You can find a detailed breakdown for the .VF Project JSON structure here.
This data can be accessed manually via the ‘Share’ menu on your project, by selecting the Export As… > Assistant File Content > Assistant File (JSON) option, or programmatically by calling the Project API.
The next step in the workflow is to transform your .VF JSON data into the required data structure. Check out the guides per NLU to understand best practice and get started.
This is the destination for your project data. With a successful workflow, you are able to pass all of your NLU Model, Logic and Response Content data created on Voiceflow into your Console, so it is ready for production, or further editing.
Updated 14 days ago