Multi-Agent Interaction

[elcade]

Describe the problem/error/question

Hello everybody!

Is it possible to have an AI Agent (Summary Agent) node interact with the user and then continue the workflow after the user decision? For example, the Summary Agent will ask: “do you have any quesrtions about the retrieved data?, for example, how many physical servers were found in the project?” Then the user can interact until satisfied, then continue with the workflow.

In the workflow all works well

1. the user starts the conversation
2. the Supervisor Agent asks a few questions and asks for validation
3. workflow continues calling APIs and retrieving a JSON file
4. here is where I’d like for the Summary Agent to ask the user (which it does) if any questions about the data
   1. the user can opt for ask the questions or say no
5. the issue occurs when the user ask a question, the Supervisor Agent receives the question but has no context of the data retrieved

What is the error message (if any)?

No errors.

Please share your workflow

I can share workflow info later if absolutely required.

Share the output returned by the last node

The last output works fine, there is a call done to MS Power Automate to create a case in Salesforce, then returns a Webhook with case information.

[details="instance information"]
# Debug info
## core
- n8nVersion: 1.118.2
- platform: docker (self-hosted)
- nodeJsVersion: 22.21.0
- nodeEnv: production
- database: postgres
- executionMode: regular
- concurrency: -1
- license: community
- consumerId: unknown
## storage
- success: all
- error: all
- progress: false
- manual: true
- binaryMode: filesystem
## pruning
- enabled: true
- maxAge: 336 hours
- maxCount: 10000 executions
## client
- userAgent: mozilla/5.0 (windows nt 10.0; win64; x64; rv:144.0) gecko/20100101 firefox/144.0
- isTouchDevice: false
Generated at: 2025-11-13T22:03:23.981Z
[/details]

Thank you so muh for your help!

[Parintele_Damaskin]

@elcade

the issue occurs when the user ask a question, the Supervisor Agent receives the question but has no context of the data retrieved

Have you tried to use same memory for all the agents node you have inline?

Connect the memory to each of the agent node, so they can share same content.

Give it a try and see… Btw, be sure you use system prompts wisely

[elcade]

Hey there!

No, I’ve never tried connecting multiple agents to the same ‘Simple Memory’ node!
Let me try this.

In regards to the “…be sure you use system prompts wisely”, is there anything you know I have to consider?
Any advise I super appreciated it.

Thanks!

[Parintele_Damaskin]

System prompt should have clear instructions, descriptions and examples.

If you wanna find more about system prompting here is a good place(not affiliated) :

Examples of Prompts – Nextra

[elcade]

I tried this and unfortunately cannot get the agents to share the memory.

[Parintele_Damaskin]

Even if you connect to same memory?

[elcade]

Yep.

Up to the first time reaching the ‘Summary Agent’, all goes well.

1. ‘AI Agent’ asks a few questions and upon user acceptance, continues calling APIs,
   authenticates, cleans JSON data and then ‘Summary Agent’ asks “Do you have any questions
   related to the data that was retrieved for your project?”

2. User answers “yes”

3. ‘AI Agent’ instead of the ‘Summay Agent’ responds.
   3.1 Sometimes with no output and somethimes just makes a summary of the previous user answers from step 1.

   
   

Should I be thinking in sub workflows?

[Parintele_Damaskin]

@elcade

Should I be thinking in sub workflows?

That would be a good approach, since you want inline 3 models to “rethink”… looks a bit cumbersome and redundant at the moment.

[elcade]

Thank you so much!