L2M

Appearance

Core Nodes ​

Input ​

  • webhook_input
  • text_input
  • system_prompt
  • user_prompt

Prompting ​

  • prompt_template

LLM / Agent ​

  • llm_call
  • agent_orchestrator
  • supervisor_node

MCP / Tools ​

  • mcp_tool

RAG / Data ​

  • rag_retrieve
  • connector_source

Utilities ​

  • simple_memory
  • switch
  • output_parser
  • http_request
  • code_execution
  • pdf_output
  • chart
  • execute_workflow

Output ​

  • output

Notes on execution ​

  • Disconnected nodes are skipped.
  • Switch nodes route branches by evaluated case.
  • Node execution status is persisted in execution history.

Output Parser (output_parser) ​

output_parser supports two independent controls:

  • mode:
    • json_schema
    • item_list
    • auto_fix
  • parsingMode:
    • strict
    • lenient
    • anything_goes

parsingMode behavior ​

  • strict: only valid JSON parses successfully.
  • lenient: tries common repair steps (python literals, single quotes, trailing commas, unquoted keys).
  • anything_goes: includes lenient and also best-effort parsing of simple key: value blocks.

Input key behavior ​

inputKey accepts:

  • path format: debug.agent_answer
  • array path format: messages[5].content
  • moustache path format: {{debug.agent_answer}}

Runtime trace ​

When mode is json_schema or auto_fix, parser output includes:

  • parsed
  • raw
  • retries
  • parserTrace:
    • strictness
    • strategy
    • confidence
    • candidateCount
    • attempts

Use parserTrace in logs to understand why a payload parsed or failed.