i created the agent with param “response_model=CustomReq”, where CustomReq is a Pydantic class.
Agno does SOMETIMES parse the JSON returned by Perplexity, and create an object of that type. The problem is that Agno doesn’t work 100%.
In debug mode, I saw that perplexity returned a string like: json { ... }. This string is well-behaved JSON. but Agno just failed to parse it, and create the object.
Is this a bug?
The problem might be caused by Perplexity returning the extra string prefix “json" and the postfix "”, which causes Agno to refuse to parse the json string, thinking it’s badly-formed?
Hi @Jie
thanks for reaching out and supporting Agno. I’ve shared this with the team, we’re working through all requests one by one and will get back to you soon.
If it’s urgent, please let us know. We appreciate your patience!
Hey @Jie could you please share the agent config so I can try to replicate the bug you’re encountering? Thanks! 
Hi, Thanks for reaching out. I’m using Agno 1.1.7 (released Feb 26). And the problem i had is with the structured output, using Perplexity (sonar-pro model).
The response from Perplexity sometimes (mistakenly) has the json string surrounded by a Markdown code fence: triple backticks, then the string “json”, then the JSON string itself, then the closing triple backticks. This is perplexity’s bug. Even if i explicitly prompt it not to include this code fence, it sometimes sends back this.
Since I use structed output, Agno would try to run json.loads() function to convert it to a Python object, and the it would choke on this input string, as it has the markdown code fence.
Is there a way to check if this markdown code fence exists, and strip it before converting it to a Python object?
Hey @Jie
We’ve shipped a bunch of bug fixes in the latest releases. Could you please update to the newest version, give it another try, and let me know if you’re still seeing the issue? Also, please share your agent config so we can replicate the issue.
Thanks