> ## Documentation Index
> Fetch the complete documentation index at: https://docs.livefaces.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Node types

> Click a node to explore its purpose, configuration, outcomes, and example.

Select any step below. The figure illustrates a route through all twelve types; a real workflow usually uses a smaller set.

NODE EXPLORER12 types · select a step

Scrollable example workflow diagram

An illustrative route through all types. Choose only the steps your workflow needs; connect alternative and unsuccessful outcomes in the editor.

<dl>
  <dt>Configure</dt><dd />

  <dt>Outcomes</dt><dd />

  <dt>Example</dt><dd />
</dl>

Link to this node ↓

## Quick reference

| Node | Use it to | Outcomes |
| - | - | - |
| [Start](#start) | Define the context supplied when the workflow starts. | `next` |
| [Say / Message](#say-/-message) | Deliver a specific message, optionally with values from inputs. | `complete`, `error`, `timeout` |
| [Conversation goal](#conversation-goal) | Hold a flexible conversation toward a defined objective. | `complete`, `error`, `timeout` |
| [Collect information](#collect-information) | Gather named, structured fields through conversation. | `complete`, `error`, `timeout` |
| [Classify](#classify) | Interpret supplied content and choose one of your named routes. | `your labels`, `error`, `timeout` |
| [Branch](#branch) | Choose a route by comparing values already present in the inputs. | `your rule outcomes`, `fallback` |
| [Run task agent](#run-task-agent) | Analyze supplied information and return structured results. | `complete`, `error`, `timeout` |
| [Authenticate](#authenticate) | Verify the participant through your configured authentication integration. | `authenticated`, `denied`, `error`, `timeout` |
| [Confirm action](#confirm-action) | Ask for explicit approval of a specific tool action. | `confirmed`, `rejected`, `error`, `timeout` |
| [Call MCP tool](#call-mcp-tool) | Invoke one approved integration tool with mapped inputs. | `success`, `rejected`, `unknown`, `error`, `timeout` |
| [Run subflow](#run-subflow) | Reuse a published workflow as one step in a larger flow. | `complete`, `error`, `timeout` |
| [End](#end) | Finish the workflow with a final status and optional output. | None |

## Configuration by node

### Start

Add Workflow inputs with names, types, and required fields. Connect next to the first step. Keep exactly one Start.

Start does not speak. Add Say / Message for an opening line.

### Say / Message

Enter Message. Add and bind any Inputs, then use Insert message variable to insert them into the text.

Use this for exact wording. A conversation goal handles an exchange with the participant.

### Conversation goal

Write Instructions covering the goal, boundaries, and when to finish. Bind relevant context and define any required Output fields. Choose Approved read tools if needed.

Keep multi-turn dialogue inside this step; the canvas does not permit cycles.

### Collect information

Describe what to ask and how to clarify missing details. Define required Output fields and types. Bind existing context so the agent can avoid unnecessary questions.

Use a later Call MCP tool when you need to verify a supplied value against an external system.

### Classify

Select Provider and Model, write Instructions, and bind the content to classify. Add at least two distinct Classification routes and connect each label. The label output uses the same allowed values.

Give labels clear boundaries and include a catch-all category when needed. error and timeout are reserved names.

### Branch

Bind the values, add Ordered branch rules, and select an input, comparison, and comparison value for each. The first match wins; connect the Fallback outcome as well.

Comparisons include equality, inequality, numeric ordering, membership, and existence. Use Classify when the category needs interpretation.

### Run task agent

Select Provider and Model, write Instructions, bind each input, and define required Output fields. Review task limits and Maximum time for the job.

Bind the evidence explicitly. Follow with a conversation or message step to explain the result aloud.

### Authenticate

Select MCP connection, Authentication challenge, and Trusted verifier. Bind the required inputs and route both authenticated and denied.

A participant claiming an identity is not verification. Use the trusted verification result and keep denied paths away from protected actions.

### Confirm action

Choose Action to confirm and describe the intended action in Instructions. Bind the same action details as the paired tool. Route confirmed to that tool and rejected elsewhere.

On a write tool, Insert matching confirmation creates the paired step. Review the action details and routes before testing.

### Call MCP tool

Choose Approved MCP tool. Map its input fields; the selected tool supplies the field definitions. Pair a write action with Confirm action and route every outcome.

unknown means the action result is uncertain. Check its status or arrange review before trying a change again.

### Run subflow

Select Subflow and an explicit published version. Map the required inputs and use the selected version’s outputs in later steps.

Subflows use the parent workflow’s voice and face. Publishing a new child version does not update an existing parent revision.

### End

Choose End status: completed, failed, or cancelled. Bind final inputs to match Workflow output fields. Keep one failed End to enable default error and timeout routing.

End does not deliver a closing line. Put a Say / Message step before it when the participant should hear one.

See [Build a workflow](https://docs.livefaces.ai/workflows/build-a-workflow.md) for a worked example, or [Data and routing](https://docs.livefaces.ai/workflows/data-and-routing.md) to connect inputs and outcomes.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.