Saltar al contenido principal

Multi-Agent

Category: Agent
Node id: multi-agent

Coordinates multiple AI agents working on sub-tasks using parallel, sequential, or debate strategies

What it does​

  • Parallel agent execution for concurrent sub-task processing
  • Sequential agent execution with context passing between agents
  • Multi-round debate strategy for consensus building
  • Per-agent model and role configuration
  • Accumulated context sharing across sequential steps
  • Concurrent execution with goroutines and mutex-safe result collection

Use cases​

  • Multi-perspective clinical analysis (e.g., specialist panels)
  • Ensemble reasoning for higher-confidence answers
  • Debate-based verification of clinical recommendations
  • Parallel data processing across specialized agents

Runtime​

  • Performance tier: slow

Input ports​

NameTypeRequiredDescription
tasktextYesThe task for agents to work on
contexttextNoShared context for all agents

Output ports​

NameTypeRequiredDescription
resultjsonNoAgent results; contains results (array), agents (count), and for debate: rounds and total_rounds
successbooleanNoWhether the execution completed without error

Settings​

Basic​

SettingNameTypeRequiredDefaultAllowed valuesDescription
LLM Endpointllm_endpointstringYesBase URL of the LLM API
StrategystrategyenumNoparallelparallel, sequential, debateMulti-agent coordination strategy
AgentsagentsjsonYesAgent definitions array (name, role, model per agent)

Auth​

SettingNameTypeRequiredDefaultAllowed valuesDescription
API Keyapi_keysecretNoAPI key for the LLM provider

Advanced​

SettingNameTypeRequiredDefaultAllowed valuesDescription
Max Roundsmax_roundsnumberNo3min 1; max 20Maximum debate rounds (debate strategy only)

Errors​

CodeDescription
ENDPOINT_REQUIREDllm_endpoint is not configured
TASK_REQUIREDNo task input was provided
AGENTS_REQUIREDAgent definitions array is empty or missing
UNSUPPORTED_STRATEGYStrategy must be parallel, sequential, or debate

Runtime notes​

  • Each agent result contains {agent, answer, status}; status may be 'error' with an error field
  • Sequential strategy passes each agent's answer to the next agent as accumulated context
  • Debate strategy returns 'rounds' (array of arrays) instead of flat 'results'
  • Parallel execution uses goroutines; result array indices match the agents config order

See Build any workflow for how to put nodes together and Ports, types and data flow for how port types connect.

¿Algo no está claro? Envíe sus comentarios sobre la documentación por correo electrónico..