HIL(human-in-the-loop) process support

[rmqrmqrmq]

Hi,
Currently, only one 'chat input' component can be used in per workflow. Is it possible to use more than one?
For example, the 'chat output' can be connected to the 'chat input' to achieve HIL(human-in-the-loop) process.
Thanks.

[anovazzi1]

hey, right now is impossible, but human in the loop is in our road map, and it will be implemented soon

[srlynch1]

agree, this is needed

[Lucky2020]

Agreed, this is a very important feature.

[juanvecino]

Hi, does someone know if there has been any updates to this?

In most flows you do not have all information at once and you would like to ask the user for the rest of the information to continue with the flow

[zhoulijing18]

现在实现了吗？ how about?

[tatwong]

is this ready?

[rahulwingify]

Hey @anovazzi1 any updates on this?

[kvrobert]

Hey, any updates on this feature? It’d be super important and really useful.

[tianzhipeng-git]

a poc implement #8427

[jeffpasqua]

Any updates on how this initiative is progressing?

[antopolskiy]

I concur this is a much needed feature

[flefevre]

Agent workflows will be accepted if we had the possibility to keep human in the loop where it is mandatory

[KeepALifeUS]

While waiting for native HIL support, here's a pattern that works today:

State-Based Human Intervention Points

state = {
    "workflow_id": "wf_123",
    "current_step": "processing",
    "requires_human_input": False,
    "human_prompt": None,
    "human_response": None,
    "agent_outputs": {}
}

# Agent decides it needs human input
if needs_clarification:
    state["requires_human_input"] = True
    state["human_prompt"] = "Please confirm the customer's address"
    state["current_step"] = "waiting_for_human"
    # Workflow pauses here

# After human responds (via API/webhook)
state["human_response"] = user_input
state["requires_human_input"] = False
state["current_step"] = "resume_processing"

Implementation Approach

1. Polling endpoint: Workflow checks state periodically
2. Webhook: Human response triggers continuation
3. State persistence: Store state in DB/Redis for long pauses

Why This Works

• No multiple chat inputs — Single entry point, state manages flow
• Any step can pause — Just update state
• Resumable — State persists across sessions
• Debuggable — Inspect state to see what's waiting

This is the stigmergy pattern applied to HIL: coordination through shared state. More details: https://github.com/KeepALifeUS/autonomous-agents