Support React Supervisor Agent that detects and calls other Agents (Agent-as-Tool) in AgentFlow v2 #5532
Description
Feature Description
I'm working heavily with Flowise AgentFlow v2 to build multi-agent orchestration similar to LangGraph (Supervisor Agent → Worker Agents → React Loop → State Updating**).
Currently, I noticed the following limitation:
❗ AgentFlow v2 does not allow an Agent to be exposed as a Tool
Which leads to several problems:
A supervisor agent cannot dynamically know which worker agents it is connected to
ToolLoader cannot load agents
React-style planning cannot call worker agents one-by-one
If I connect multiple agents into a single Agent node, Flowise calls all of them at once
Conditional Agent does not solve the issue (no planning, no react loop, no state updates)
Using a plain LLM node requires manual prompting + JSON planning, which is very hacky and breaks the purpose of AgentFlow
What I'm trying to achieve
I want a structure like this (same as LangGraph Supervisor):
Supervisor React Agent
→ dynamically reads list of available worker agents
→ plans which agent to call
→ calls exactly one worker at a time
→ receives observation
→ continues ReAct loop
→ final answer (JSON or text)
Worker Agents:
should be callable individually
should expose a description / capability
should behave like “tools” for the supervisor
should not all be invoked in parallel
Current Workaround
Right now the only possible workaround is:
Wrap each Agent Worker inside a Custom Tool
→ Tool does HTTP POST back to Flowise API endpoint of that agent
→ Supervisor React Agent calls these proxy-tools
This works, but:
it's very manual
requires custom code for every agent
breaks the simplicity of AgentFlow
duplicates the agent definitions
loses the “no-code” experience
Feature Request: Support Agent-as-a-Tool in AgentFlow v2
This means:
Allow an Agent node to be marked as “Expose as Tool”
Give it a tool description
ToolLoader & Agent nodes can load these Agents as callable tools
React Agent can:
automatically see available agents -> plan -> call 1 agent at a time-> perform ReAct loops -> update flow state
This would enable true multi-agent orchestration in Flowise, and fully unlock patterns similar to LangGraph supervisor+workers.
Why this feature matters
True multi-agent systems require a supervisor → worker calling mechanism
AgentFlow v2 currently lacks the ability for agents to call other agents
Building large agentic flows becomes extremely difficult without this pattern
Current workaround (Custom Tool wrapper) breaks the no-code nature of Flowise
Competitors (LangGraph, CrewAI, AutoGen) support this architecture natively
Feature Category
New Node/Component
Problem Statement
No response
Proposed Solution
No response
Mockups or References
No response
Additional Context
No response