Voiceflow Data Structure

Learn the structure of a Voiceflow project file

In this guide, we will cover the composition of the Voiceflow Project file (.vf file).


In this guide, the term "project" is used to refer to an Agent.

It's broken up into 3 parts: Project, Version, and Diagram, each of which will be covered here.


Voiceflow project file export


Project metadata, associated with a particular team/workspace

  • _id unique ObjectId Identifying this particular project
  • name name of the project
  • teamID team/workspace that this project belongs to, hash-id string
  • devVersion ObjectId pointer to the main working version on the project
  • platform the channel of the project (alexa, google, chatbot, twilio, etc...)
  • platformData project metadata for the particular platform type
  • members distinct metadata for members working on the project -(configurations or settings that different members might have)

...other fields with more additional metadata for various functions


A project can contain many versions, the devVersion is the main working version. Versions can be used for staging as well as backups.

  • _id unique ObjectId Identifying this particular version
  • name name of the version (i.e. "backup save before change X")
  • projectID project that this version belongs to
  • rootDiagramID ObjectId pointer to the home/root diagram. This is the starting point.
  • platformData version metadata for the particular platform type
    • intents
    • slots
  • variables array of global variable names
    ...other fields with more additional metadata for various functions

Diagrams (Topics and Flows)

Version: A version contains many diagrams, with the rootDiagram being the entry point during runtime.
Can find all diagrams of a version searching against the versionID index.

These might also be referred to as topic, or component. Both are types of diagrams. An older name for diagrams is "flow"

Topics: topics are all on the same level. Can be thought of as extensions of the home diagram. On runtime they blend together like one giant diagram.

Components: components are modular, reusable pieces. You can think of components similar to functions in programming. When a component gets called, it is added to the "call stack", and pops off when it is finished.

  • _id unique ObjectId Identifying this particular diagram
  • name name of the diagram
  • type type of diagram: TOPIC or COMPONENT
  • versionID version that this diagram belongs to
  • nodes dictionary of all the nodes in the diagram, referenced by their nodeID. This is where most of the data lives.
  • variables array of diagram-level variable names (local variables)
    ...other fields with more additional metadata


Every node has:

  • unique nodeID (only needs to be unique per diagram)
  • type canonical type of node
  • data that houses metadata for the node.

blocks or steps are types of nodes


A block is comprised of steps, and usually, the last step in the block will have ports. There are "implicit" ports between steps within the same block.

A line coming out of a port can be connected to another block or step.

Special steps might have no ports, or can not be connected to (they are a triggered event)

Markup, images + text on canvas with no functional purpose, are special types of nodes.


Blocks contain a steps property in data that references their children steps, in this case, a speak step + a choice step are always of type: block.


Below is what the above "Greetings" block looks like in JSON form.

"619a67a19eb63ecb29430946": {
  "nodeID": "619a67a19eb63ecb29430946",
  "type": "block",
  "coords": [
  "data": {
    "name": "Greetings",
    "color": "blue",
    "steps": [



See what the "choice" step looks like in JSON form below:

// simplified version
"619a67b59eb63ecb2943094f": {
  "nodeID": "619a67b59eb63ecb2943094f",
  "type": "interaction",
  "data": {
    "choices": [{
        "intent": "VF.YES", "action": "PATH"
        "intent": "VF.NO", "action": "PATH"
    "ports": [{
        "type": "else",
        "target": null,
        "target": "619a67bc9eb63ecb29430959",
        "target": "619a67bf9eb63ecb29430967",

type is the canonical type of the step. There will always be ports under the data, which target another node.

Everything else in the data is metadata dependent on the particular step type.

Required Fields

Below are the required fields in the VF file that are needed for successful import.

  "project": [
  "version": [
  "diagrams": {
    "[DIAGRAM ID]": [