Workflow node reference

Choose a node for the task

The current editor exposes 28 node types. Add a node, configure its fields, connect its ports and test its inputs/outputs before publishing.

Start points, AI and data

Declare inputs at the start or trigger node. Select models/instructions for LLM and agent nodes, and a base/query for retrieval. Map the answer or output fields at the end.

1. Start
2. LLM
3. Knowledge retrieval
4. Answer
5. End
6. Agent
7. Data source
8. Webhook trigger
9. Schedule trigger

Understand requests

Define clear classes and connect each classifier branch. For parameter extraction, define names, types and expected information; test missing-data cases.

1. Question classifier
2. Parameter extractor

Conditions and loops

Cover conditional branches and defaults. A decision table returns values; add If/Else to branch on them. For iteration/loops, define inputs and termination conditions.

1. If / Else
2. Decision table
3. Iteration
4. Loop

Transform data

Select the correct inputs, types and outputs. Test templates, variable aggregation/assignment, lists and document extraction on small data. Code nodes require an operations-configured sandbox.

1. Template
2. Variable aggregator
3. Variable assigner
4. List operator
5. Document extractor
6. Code

Integrations and outputs

Configure permitted HTTP destinations, tools or channels. For knowledge writes, select the destination base and inspect ingestion status. Check authentication, schemas and egress policy.

1. HTTP request
2. Tool
3. Send via channel
4. Write to knowledge

Waits and approvals

Configure resume time/events and required reviewer input. When a run stops here, inspect its waiting state and approval queue before treating it as a failure.

1. Wait
2. Wait for event
3. Human input