Files and agents

Atwork's AI agents read the files you attach in a chat and the files on an item in a workflow, and they can create PDF, Word, Excel and CSV files for you. This page covers what an agent reads, how much of each file, and what it can make.

Attach a file in a chat

Click the paperclip, Attach file, in the message box of any agent chat: Ask Dana, a chat with one of your agents, or a sheet's Insights. Each click adds one file, so click again to add another.

• The file shows in your message as a chip with its name.
• The agent reads up to 5 files per message. Attach more and it reads the first 5, so send the rest in your next message.
• You can attach a file and ask about it in the same message: "Pull the deadline, the deliverables and the approver out of this brief."

What the agent reads

Agents read documents and text files as text, and see images when their model supports vision.

Pictures inside a PDF, Word file or deck aren't seen; only the text is read. Attach the images that matter as separate files.

How much it reads

An agent reads files up to 10 MB each, and only the start of very long ones.

• 10 MB per file. Bigger files are skipped.
• 50 pages per PDF. Pages after the 50th aren't read.
• 30,000 characters per file, and 60,000 across the files in one message. When a file is longer, the agent is told it read only the start, so it can tell you.

Images and models

An agent sees an image only when its model supports vision. Dana runs on a model with vision.

• Some models can't see images, for example DeepSeek V4 Pro and Grok 4.3. In a chat, an agent on one of them gets only the image's name, says it can't see it, and suggests switching the agent to a model that can. In a workflow, the image is skipped.
• Some models take fewer image types. Gemini 2.5 Pro and Gemini 2.5 Flash don't take GIF.
• A saved agent's model is set in Settings → Agents.

Files in workflows

In a workflow's Agent step, turn on Read files attached to the item on the Task tab, and the agent reads the files in the item's File List fields.

• Up to 5 files per run, 10 MB each (images 3.75 MB, on a model with vision).
• The agent reads the files on the item the step receives. Files on other rows, such as a contract on the client's row, aren't read.
• Start the workflow once the file is on the item. Item Created runs as soon as the row exists, so if people add the file afterwards, use a Field Changed trigger, for example on Status.

In Executions, a file the agent skipped shows under Not read: with the reason:

• not found in this workspace
• over 10 MB
• type not supported
• no text layer
• the model can't see images
• couldn't be read

The rest of the step is covered in The Agent step in workflows.

Files agents create

Ask an agent for a document or an export, and it creates the file and adds it to the conversation as a file chip.

• Documents: a PDF or Word document from what the agent wrote, such as a kickoff summary or a client report.
• Exports: an Excel (.xlsx) or CSV file of rows from a sheet, up to 10,000 rows. When more rows match, the agent asks you to narrow the export with a filter.
• Size: up to 10 MB a file.
• Preview: click the chip to preview a PDF, or the first 100 rows of an Excel or CSV file, for files up to 5 MB. Word documents have no preview; use Download.
• Where it lives: the file stays in the conversation and isn't attached to any record. Download it and add it to the row where it belongs.

Which agents can create files

• Dana can, out of the box.
• Your own agents: in Settings → Agents, open the agent, and under Capabilities set Files (PDF, Word, Excel, CSV) to Read & write.

Reading files you attach doesn't depend on this setting.

Limits at a glance

Frequently asked questions

Can an agent read a scanned PDF?

No. A scan is a picture of each page with no text layer, so there is nothing to read, and a workflow run shows "no text layer". Ask for a PDF exported from the original file, or attach photos of the key pages as images to an agent whose model has vision.

Why did the agent ignore my image?

Its model may not support vision, the image may be over 3.75 MB, or the type may not be one the model takes, such as HEIC, or GIF on Gemini 2.5. Save it as a PNG or JPEG under 3.75 MB, or switch the agent to a model with vision.

Can I attach more than 5 files?

You can, but the agent reads the first 5 in each message. Send the others in your next message.

Where are the files an agent creates saved?

In the conversation, as a file chip. They aren't attached to any record, so download the file and add it to the row where it belongs.

Does reading files use the AI budget?

Yes. A file's text is part of what the agent reads, so a long file uses more of your workspace's AI budget than a short message. See The AI budget.

Related: The Agent step in workflows · Meet Dana, your AI assistant · Create an agent and give it knowledge · The AI budget