Workflow Nodes Reference
Look up node types, field defaults, runtime limits, and validation rules
Workflow definitions currently use schema_version: 2. The following fields come from the backend node configuration schemas and the frontend editor registry.
Node Fields
| Node | Core fields | Default/Limit |
|---|---|---|
| LLM | model_id, prompt, temperature, max_tokens, top_p | temperature=0.7, range 0-2; top_p range 0-1 |
| Agent | agent_id, message | Agent must exist and be accessible |
| Knowledge Base Retrieval | knowledge_base_id, query, top_k, score_threshold | Top K default 5, 1-100; threshold default 0.3, 0-1 |
| Conditional Branch | expression, true_output, false_output | The two outputs default to true/false |
| Loop | array_expression, item_variable, index_variable, max_iterations | item=item, index=index, max 100, 1-1000 |
| Code Execution | language, code, timeout_seconds | Python/JavaScript; default 30s, 1-300s |
| HTTP Request | method, url, headers, body, timeout_seconds | GET default; timeout default 30s, 1-300s |
| Tool | tool_id, arguments | Arguments mapped by expressions |
| Sub-workflow | workflow_id, input/output mapping, timeout, failure policy | Timeout default 300s; fail_on_error=true |
| Delay | seconds | 0-3600 |
Editor Nodes
The current UI also registers user input, trigger, media generation, iteration container, iteration start/end, loop start/end, template transform, file-to-URL, variable aggregation, variable assignment, parameter extraction, question classification, output, and comment nodes.
Validation
Pre-publish checks reject nonexistent variable references, empty prompts, unselected models/tools/knowledge bases, missing input sources, missing output nodes, incomplete branches, and incomplete containers. The start node cannot be deleted.
Data Format
nodes and edges use the React Flow structure; viewport defaults to {x: 0, y: 0, zoom: 1}. Version 2 supports native objects and arrays, and nodes can save inferred schemas and output type specifications.
How is this guide?