Chat Attachments
Let end users upload files in a chatbot conversation so your agent can see, read, and reason over them.
Chat Attachments
Chat attachments let end users hand your agent a file - a screenshot, a PDF, a CSV, a text dump - right inside the chat. The agent then sees that file and can answer questions about it, summarize it, extract data, or use it as context for whatever it does next.
This is what makes flows like:
- "Here's a contract. Tell me what's risky."
- "Here's a screenshot of the error. What's wrong?"
- "Here's our pricing CSV. What's our average margin?"
...actually work, instead of asking the user to copy-paste everything into the chat.
Supported file types
FormWise accepts the following kinds of files:
| File type | Extensions | How the agent uses it |
|---|---|---|
| Images | .png, .jpg, .gif, .webp, etc. | Passed to the model's vision - the agent can see and describe the image |
| PDFs | .pdf | Text is extracted and indexed for retrieval - the agent searches the doc as it answers |
| Spreadsheets / CSVs | .csv | Text is extracted and indexed for retrieval |
| Plain text | .txt | Read inline if small; otherwise indexed for retrieval |
Limits - Plain-text files under 200 KB are read inline so the agent has the whole file in front of it. Larger files (and PDFs / CSVs) go through retrieval, with a per-file cap of 25 MB.
How end users attach a file
Every chatbot Flow shows an attach button in the chat input bar - there is no setting to enable or disable it.
- The end user taps the + button (paperclip / attach icon) in the chat composer.
- A menu appears with two choices:
- Upload file - pick a file from device.
- Previous uploads - pick from files this end user has uploaded before in any of your Flows.
- The file uploads and appears as a chip above the chat input.
- The user types their message and sends. The agent receives the attachment with that turn.
[Screenshot: attach menu with Upload file and Previous uploads options]
Previous uploads
End users don't have to re-upload the same file every time. Files they've uploaded to your Flows stay accessible to them under Previous uploads - handy when someone wants to ask follow-up questions about a doc days later.
Previous uploads are scoped to the end user, not shared across users.
Behind the scenes
Different file types take different paths so the agent gets the most useful version of the content:
- Images go straight to the model's vision input. The agent literally sees the image and can describe details, read text in the image, identify objects, etc.
- Small text files (under 200 KB) are pasted inline into the agent's context. Useful for code snippets, logs, short transcripts.
- PDFs, CSVs, and larger text files (up to 25 MB) are run through the same retrieval system used by your Notebooks. The agent then searches the doc as it answers, returning grounded responses.
You don't have to choose - FormWise picks the right path per file.
Tips
- Set expectations in your persona. If your Flow handles attachments, mention it in the agent's greeting: "You can drop a screenshot or PDF in the chat and I'll work from it."
- Big PDFs work, but think before you paste. A 500-page contract is fair game - but the agent will still need to find the right passage. Asking "What are the renewal terms?" works better than "Tell me about this PDF."
- Combine with artifacts. A common flow: end user uploads a doc, agent reads it, agent produces a polished summary artifact the user can edit and download.
- Use Notebooks for things you always want. If a file is reference material for every conversation, add it to your Notebooks instead of expecting users to upload it each time.
Limitations
- File types other than images, PDFs, CSVs, and plain text are not accepted.
- The 25 MB per-file limit applies to PDFs, CSVs, and large text files. Larger files won't upload.
- Image understanding depends on the model you pick. If you've chosen a model without vision support, image attachments won't be processed - swap to a vision-capable model in the Agent node settings.
- Voice mode and attachments work together: end users can attach a file, then talk about it.