human in the loop control

[cchiaramelli]

Is your feature or enhancement related to a problem? Please describe.

Would it make send to build human in the loop control feature for Chatwoot?

If so, do you have ideas on how it should be, or willing to accept suggestions?

Describe the solution you'd like

Still understanding if its worthy

Describe alternatives you've considered

No response

Additional context

No response

[pranavrajs]

@cchiaramelli Could you please provide more context?

[cchiaramelli]

Sure

I've been using Chatwoot for Conversational AI use case, where messages are answered automatically by a chatbot.

Default workflow is:

• When conversation is Pending, the AI agent responds to incoming messages.
• When conversation is Open, the AI agent ignores the conversation, and a human agent is expected to respond via the UI.
• The AI agent can also transfer the conversation to a human at any time by changing the conversation status.

This works well for regular AI-to-human handoff. However, we also have a Human-in-the-Loop use case where the conversation needs to be transferred to a human, the human agent may exchange messages with the user and enrich the conversation with a specific “result”, and then the conversation should be returned to the AI agent so it can continue with the additional context.

For now, we solved this using custom attributes. At the end of the loop, the human agent sets the result in custom attributes created with a specific prefix, such as hitl_. Our application identifies changes to those attributes, sets the conversation status back to Pending, and provides the additional context to the AI agent.

This works, but it requires training human agents and is prone to failure, since the HITL flow depends on indirect/manual actions. I know this is an AI-specific use case, but it may be useful for other workflows too. My initial thought is that, when enabled for a tenant, Chatwoot could provide UI and API support for Human-in-the-Loop workflows, such as:

• Making it clear to the human agent that the conversation is part of a HITL process, not just assigned to them normally.
• Allowing the human agent to explicitly complete the HITL process and provide structured “result” data.
• Making previous HITL process results visible in the conversation context.

Would something like this be aligned with Chatwoot’s roadmap?