Attachments: read, anonymize and edit
Who this is for: whoever manages the account.
What it is for
Customers send photos and documents. An agent can read them to extract data, and also cover sensitive information before forwarding them.
Reading attachments
With read_attachment the agent reads PDF, Word, Excel, CSV, text and photos of documents. Photos are read with OCR (text recognition) that runs on TalkAI's own server.
Editing images and PDFs
With edit_media the agent can redact, crop, retouch, combine, rotate, convert or turn an image black and white. It can also redact PDFs.
- In a PDF, redacting is destructive on purpose: pages are turned into images, the areas are covered and the file is rebuilt. The result no longer has searchable text, so what was covered cannot be recovered.
- Redaction, crop and retouch edits on images are done by a generative model: it returns the already edited image. That is why it is worth checking the result before using it, especially on documents where every detail counts.
ℹ️
Note:The Image editor template already comes with rules about what may be edited. See Templates.
How to set it up
➡️
Step:Create the agent from the Document reader or Image editor template, or add the tools by hand.
➡️
Step:Define in the prompt which documents it expects and what data to extract or what to redact.
➡️
Step:Try it in the test chat by attaching a real file. See Test chat and activation.
Common mistakes
- Blurry or cropped photos: ask the customer for another one.
- Not reviewing an edited image when the redacted data is critical.