Product Feedback: Custom agents extracting data from attachments

Update: this apparently only happens with inline attachments

I’m also having difficulties getting Custom Agents.  Is this a supported feature? 

When I ask it to parse a PDF using OCR or other tools I get this as a response in the ticket: 

Parsing Notes / Errors

• The PDF was successfully downloaded (217,403 bytes, content type: application/pdf).
• The PDF content was acknowledged by the extraction tool but did not return raw, human-readable OCR text suitable for structured field-level parsing. The tool returned a generic "[PDF content extracted]" placeholder rather than the actual text content of the document.
• Fields such as Distributor Number, Distributor Address, Primary Contact, Alternate Contact, and Warehouses could not be confirmed or extracted due to this limitation.
• The PDF likely contains scanned/image-based form fields that require a dedicated OCR engine (e.g., Tesseract, Adobe OCR) to fully parse.

When I ask the agent to parse a similar document in a DOCX format using OCR or other tools I get this as a response in the ticket:

⚠️ Parsing Errors / Notes

The attachment (SpikedAde VIP Commitment Form_ANM.docx) is a Microsoft Word (.docx) file, which is a ZIP-compressed XML archive. Only two fields could be reliably confirmed from the document's filename:

• Route Accounting System: VIP — derived from "VIP Commitment Form" in the filename.
• Supplier Name: SpikedAde — derived from "SpikedAde" in the filename.

The remaining fields — Distributor Name, Distributor Number, Distributor Address, Primary Contact, Alternate Contact, and Warehouses — are embedded within the compressed XML body of the .docx file and could not be fully extracted with the available parsing tools.

​@Anton21 I had the same issue originally as you with the fake URL.  I had to explicitly tell the agent not to hallucinate or make up any information as it would hallucinate the URL it thought it was supposed to use to read the attachment.  

I’ll add to this, I’m trying to get it to extract SAML metadata info from an attachment or a link. I cannot get it to do either.. it just says “Note: The SAML metadata URL or XML attachment could not be fetched or parsed in the prior step. Please manually retrieve the metadata from the ticket and populate the above fields.”

We have someone from the engineering team currently investigating this and will provide an update here as soon as possible.

Hi everyone, 

I'm the Engineering Manager for Custom Agents. Thank you all for reporting this. Your details were very helpful in identifying the root cause. Our team has found the issue affecting attachment parsing including the host-mapping mismatch for inline/email attachments and the limitations with PDF extraction and we currently have a fix that is being tested. We expect to roll it out within the next week. I'll post an update here once it's live.
Additionally, DOCX support is coming soon.

If you run into any additional edge cases in the meantime, please keep sharing them in this thread.

Thanks Setuc!

Does it support XML out of the box as well?

Hi Ohmygosh

we don’t have any support for XML at the moment. What will be your use case for XML? 
 

Setu

Hello,

We receive XML files for SAML integrations. We get easily 80 of these a year.  We want to use zendesk to extract the metadata in those files.

Hi Ohmygosh,

 Let me work with the product team to work out on how we address this requirement to process XML data. Could you please work with your account team to setup a meeting with us, so we can understand the examples a bit more, as i think there could be a way we can help you right away (depending on the complexity of the XML file). Ask your account team to reach out to the custom agents team or post on our internal slack (ask-custom-agents) and we can take it up from there. 

Setu Chokshi

Thanks! I’ve sent the request off to our account team.