This might just be a me problem, but i think i just discovered a problem with the latest AI Agent node.
When Using the latest version of the AI Agent node, it simply does not output the token data.
I have of course turned on Return Intermediate Steps for both. Please come with suggestions that might help.
achamm:
im not sure what’s going on but i think it’s possible that the node is just skipping over the token data or something similar dont wanna mess with n8n docs without more info about how they handle token outputs anyway this is how it looks for me:
let me try to debug it by checking if there are any errors in the node configuration or if you could provide some more context about your workflow and what’s happening when you run it?
You are an email classifier for a staffing/booking company. Analyze emails and extract structured data.
Respond with ONLY valid JSON, no markdown or explanation:
{
“is_work_related”: true/false,
“is_deal”: true/false,
“confidence”: “high”/“medium”/“low”,
“reason”: “brief explanation”,
“extracted_data”: {
"event_date": "YYYY-MM-DD or null",
"event_time": "HH:MM or null",
"location": "string or null",
"headcount": number or null,
"duration_hours": number or null,
"client_name": "string or null",
"notes": "any other relevant details or null"
}
}
If the email appears to be a reply (subject starts with “Re:”, “RE:”, “Fwd:”, or the email has an In-Reply-To header), you MUST use the “Check Thread in Supabase” tool BEFORE classifying.
Call the tool with:
{
“thread_id”: “”,
“sender_email”: “<sender’s email address>”
}
If the tool returns matching deals:
is_work_related should be TRUE (it’s part of an existing conversation)
Include the matched deal ID in your response under “matched_deal_id”
If the tool returns no results and it looks like a reply:
Use your judgment based on the email content
Note in the reason that no thread match was found
Rules:
is_work_related: true for booking requests, business inquiries, event planning, client communications
is_work_related: false for spam, newsletters, personal emails, automated notifications
is_deal: true only if this is a new or ongoing booking request
Only extract data that is EXPLICITLY mentioned - use null for anything not stated
For dates: parse relative dates (tomorrow, next Friday) based on current date context
confidence: high = clear booking request, medium = probable, low = uncertain
Classify this email and extract booking information:
{{ $(‘Prepare Email Data’).item.json.email_summary }}
Thread Information:
If this appears to be a reply to a previous email thread, use the Check Thread tool first.
There shouldnt be any issues with the configuration of the nodes, or the dataflow. Everything is working as it is supposed to. The only problem is that i have to use an outdated Agent node to actually get the token data, which is undesierable.
Yea I noticed that to it isnt returning the token data. I can see it in logs, but not from the agents output.
@Mathias_Brenden I think your best bet is to create a suggestion in the n8n suggest part of the forum and keep using the older node, sorry I don’t have the best solutions for you.
No problem, i assume this will be fixed sometime soon anyways. Thanks for your time!
Your welcome! I would mark one of our replies as the solution so the post is closed
Hi @Mathias_Brenden,
Do you require the token data in the output of the agent? Whats is your use case for getting back the token data?
You can still see the tokens used in the logs tab:
