You can attach documents, spreadsheets, images and text files to your conversations with the Modeler, the Analyst and Custom Agents. This article covers how to attach files, which file types and sizes are supported, how the automatic security check works and what to do when a file fails the check.
How to attach files to AI Chats
Open an AI Chat.
In the chat input field, select the plus (+) symbol.
Select Attach a file and browse for the file on your computer.
It must be one of the file types listed in the next section.
You can also add files to AI Chats from Pigment Files. For more information, see Files in Pigment.
File types
The Agents accept the following file types:
PDF, PPTX, DOCX. The Agents can analyze text, tables and images, using spatial awareness to identify relationships between elements.
CSV, XLSX. The Agents can view data, formulas, workbooks and links between workbooks. They can import data into a Pigment Block.
PNG, JPG. The Agents can read charts, screenshots and photos of documents.
TXT, MD. The Agents can read plain text, such as notes, extracts and documentation.
File size limits
You can attach up to 10 files at a time.
Each file must be less than 32 MB, or 4 MB for image files.
The total size of an upload must be less than 100 MB.
ℹ️ Note
Files are automatically deleted after 14 days.
Files sent from Pigment Files must also meet the above criteria. For more information, see Files in Pigment.
Automatic security check
All files under 4 MB in size automatically undergo a security check for text that reads as instructions to an AI Agent.
How files fail the security check
Before an AI Agent reads a file, the security check scans it for text that could instruct the Agent to expose or change data it should not access. If the check detects this type of text, the file fails the security check.
This check runs only on files under 4 MB in size. Files over 4 MB are not checked.
Troubleshoot a failed security check
A file can fail the security check even when the content is safe. This can happen when a file contains text that resembles instructions for an AI Agent.
Review the file for the following content:
AI prompts
AI policies or instruction templates
meeting notes that quote a chatbot
hidden text in a downloaded PDF
phrases such as “ignore previous instructions” or “you are an assistant”
If the file still fails after this, submit a Support ticket.
Security check output
For files under 4 MB
Attaching a file under 4 MB results in one of two outcomes:
The file is attached. It passed the check and is ready for your Agent to use.
The file is not added. Pigment displays why it failed the check:
If it found text that reads as AI instructions, attaching the same file again gives the same result. Troubleshoot the file and try again.
If the upload itself failed, check the file type and size limits and try again.
For files over 4 MB
The file is too large to check. Pigment allows you to upload it without running the check. Select Continue with Caution to attach it anyway, and treat its contents as unchecked.
What the check does and doesn’t do
Pigment checks attachments using Model Armor, a Google Cloud service that screens content passing between a user and an AI model. Pigment uses only its prompt injection and jailbreak detection filter, which identifies the instruction-like text described above.
🛡️ Warning
The security check does not look for sensitive data, malicious links, malware or harmful content.
Do AI Chat attachments count toward file storage?
No. Files you attach to AI Chats are stored separately from assets you hold in Pigment Files. They do not count toward your workspace file storage and they do not appear in your file storage usage. For more information, see Manage File Storage Limits.