I have a workflow that responds to emails based on a payload from a webhook. It created a response that was partially from the payload, but included other information. How do I find out where that other information came from?
hey rick! happy to take a look for you. can you please share the task URL?
How to share your Lindy task URL: 1. Click your Lindy agent → click “tasks” 2. Find the task you need help with 3. Click the specific task (you should see the full execution) 4. Copy that page URL from your browser 5. Paste the URL in your support request Note: The URL should include “task” followed by a number Screenshots are also helpful!
Let me know if this is what you're looking for:
The step is Send Reply. The Body field has this: Use [http output] to write a response to the customer. Make note of an email address mismatch but answer questions anyway.
not sure what the additional info here is but is it possible that the LLM just learned this through the available data online? Like would claude be able to provide that information?
In this case, the web hook only provides order information about the specific contact who sent the email. The part of the answer regarding how to sign on to the other web site is not provided at all. I guess that's my question. Does the AI have the ability to look at other sources for the answer or is it limited to the source that I provide, in this case the web hook response?
yeah no it’s limited to the information you provide it with, previous actions in the current task and it’s general training data (the data the LLM provider trained the model with) 🙂
Marvin A. I've finished building my email responder workflow, using my email address during testing. What I need to do is have it monitor two other email boxes instead of mine. I tried the Add Account option on the email received step, but it won't authenticate. I had the user create a Lindy account (which I didn't think was necessary, and is going to cost an extra $19/month). Is it possible to add other addresses to the email received step, and can that be done without creating a new lindy account (that could get really expensive if I get more customer service agents)?
Thanks. I'll have to get back with one of my advisors to try and authenticate. But, what you're saying is that I don't need to add those users to the workspace if all I want to do is have the Email Received step use their G Suite address?
But, from what I remember when we tried this morning, when I tried to authenticate that account, it wanted to create a Lindy account.
That's correct. Do you mind mind sending a screen-recording or screenshots of what it's showing you when this happens? That will help me see what's going on exactly.
I'm taking some screen shots. I'll include them in the email.
Great, thanks!
Perfect! Once that’s resolved, you should be good to go.
I have a follow-up question about my workflows and credit usage. I implemented two copies of the email responder workflow that we've been talking about. I noticed today that a lot of credits had been used. I checked in the Billing section where it lists the top 10 tasks and I found two email chains, one that used 450 credits and one that used 125. They seem to stem from an AI step that is supposed to check the body of the email for a phone number. The two in question used 37 credits to find the phone number while it normally is 1-2 credits. Each of the suspect emails had large documents attached to them. What was even more peculiar is that subsequent steps, such as marking the email as read, setting a label and replying are also being charged 37 credits. I've updated the workflow to tell the AI to not process attachments when getting the phone number but I haven't seen a new email come through for it. Does this sound normal? Why would basic Gmail tasks incur such high credit usage because of a prior AI step?
Hey Rick -- Do you mind sending the Task URL for these? If they are processing a lot of into (like a ton of HTML, or images) it could take up a lot of credits. What you can do is filter them out in the first step to prevent big files like that from coming through
Hold on. I'll get the URL...
I thought about the filter, but I don't want to exclude the email if it has an attachment. I just want the workflow to ignore the attachment (if that is what is causing the credit usage).
Did you have a chance to look at the Gmail steps for that task? It does appear that it is an outlier in terms of credit usage, but I'm still concerned that an email with a large attachment will use many credits on what appear to be simple steps.
Hey Rick! Apologies about the delay - Just took a look. Yeah, it looks like in the task you sent, there is a pretty large PDF. It took 37 credits just to summarize it and find the Phone number. It's a pretty hefty job to convert a PDF like that into text to find a phone number
I can understand that. If you look at the subsequent steps: Mark Email As Read (37), Phone Number Check (37.1), Response Conditions (37.1), Add AI Issue Label (37.2). Is it summarizing the PDF in each of those steps? Really, I don't need it to convert any attachment at all. We would only be using them to move to the customer's account in our CRM. Is there a way that I can have it ignore any attachments when searching for the phone number but continue to search the body of the email?