While following the OutSystems AI Agents journey, I ran into two errors when testing the third agent (CallAgent03_Underwriter). On My App i see that the first 2 agents are executed corretly:
My Workflow looks like this:
And this is what i see on the Instances List:
When i go to the details, the first error that i see as soon i fill the form on the app (to trigger the events)
1. Failed to call AI model.Status code: TooManyRequests. You've reached the current rate limit for the Amazon Bedrock AI model: 3 requests per minute. The limit will reset in 1 minute. Please try again later or consider using a different model. (OS-ABRS-FM-42900)As you may know the exercise proposes the use of at least 3 agents; I presume they are used sequentially. Therefore, there seems to be a problem here due to the calls being made too close together...so i wait 1 minute and then i hit the Retry button on the Instance Detail page...when i do that i get a second error:2. Failed to call AI model. Status code: BadRequest. The Amazon Bedrock AI model provider returned this message: "BedrockException - {"message":"A conversation must start with a user message. Try again with a conversation that starts with a user message."}" (OS-ABRS-FM-40001)
You can see the error also here:
I don't have any errors in my app; in fact, because of this, I completely redid the lesson to recreate the agent... but I'm getting the same result.Although I haven't yet had my first opportunity to join a team to work with OutSystems, it's a technology I appreciate and enjoy learning. I would very much like to be able to complete this exercise and perhaps obtain the Certificate and do it successfully.
If anyone can help me, I would be very grateful.André Matias
Hello @André Campos Matias,
Please check if the UserMessage content in the BuildMessages action is set to a non-empty string. I have attached a screenshot from my app for your reference.
Settings.UserPrompt has been set to a non-empty string literal mentioned in the Exercise: Based on the input, provide a decision to approve or reject the loan, with an explanation. The complete response should be less than 750 characters. Instructions for your OUTPUT: The AI agent returns the defined structured output similar to a JSON with the following keys: A key named `Text` with a string value containing the analysis (in Markdown format, concise, and user-friendly). A key named `Decision` with a string value: either `Approved` or `Rejected`.
Thanks and Regards,
Tushar Godbole
Hi @Tushar Ram Godbole and thank you so much for your kind help and taking the time to consider this issues.Thank you very much for your suggestion. I checked in Agent 3 and the association was done correctly. However, this made me also check the two variables in the settings. And here was a problem: for some reason, the text in the prompts was both empty. I tested the process again from the beginning; Agent 1 and 2 run correctly, but Agent 3 continues to give this error...Status code: TooManyRequests. You've reached the current rate limit for the Amazon Bedrock AI model: 3 requests per minute. The limit will reset in 1 minute. Please try again later or consider using a different model. (OS-ABRS-FM-42900)
If I wait 1 minute and retry the workflow, it now executes without a problem. So your suggestion helped to unlock part of the problem. Thanks!!I ran the form again, and then I understood. The goal of the exercise is for the user to keep pressing the refresh button to get updates. When this error of excessive calls occurs, it's not shown to the user... we only have that information in the workflow editor... so if the user waits (as they normally would), it doesn't end up being an error itself.
Thank you again for your help.
Hello @André Campos Matias ,
You are welcome!
Hi @André Campos Matias ,
Looks like you are making too many request to the AI model, please try to wait for 10 minutes before making another request, also keep in mind you only got 1000 calls as trial model, use them carefully.
For UW agent are you passing the value of user prompt to the model, can you please validate all the steps again carefully, sometimes, we skips the steps in rush.
Also, try to debug you agent and see where the problem is using the below documentation if you are new:
https://success.outsystems.com/documentation/outsystems_developer_cloud/building_apps/build_ai_powered_apps/agentic_apps_in_odc/debug_agentic_apps/
Regards,
Manish Jawla
Hi Manish, thank you so much for taking time to reply.I'm following this tutorial...https://learn.outsystems.com/training/journeys/build-agentic-powered-app-3411/build-the-underwriter-agent-exercise/odc/10819
The tutorial doesn't inform us about limitations, etc... and if I made any calls, it was 10 or 20 times... while I was doing the tests.
That's why I find it strange.
here is the link about limitation:
https://www.outsystems.com/product-updates/odc-ai-model-trials/#:~:text=With%201%2C000%20free%20trial%20requests
Thanks,
Thanks Manish for your kind help.Correct. That link confirms that the model allows 1000 requests. Which is even stranger because, following the tutorial so far, I've probably only made 10-20 requests, if that.
I find it strange that this bug exists and that nothing about it is mentioned in the tutorial.