File Context

You can attach files to your messages to give the agent additional context. Drag a document into the chat, and the agent can read, summarize, or analyze its contents. NimbleBrain extracts text from common formats and includes it in the conversation; images are passed straight to vision-capable models.

How to attach files

Three ways to add files to a message in the web UI:

• Drag and drop — drag a file from your computer onto the chat input
• Paste an image — copy an image and paste it directly into the input (Cmd+V / Ctrl+V)
• Click the attachment icon — use the paperclip button next to the input to browse

Attached files appear as chips below the input. Remove a file by clicking the X on its chip before sending.

Supported formats

Format	Extensions	What the agent sees
PDF	.pdf	Extracted text from all pages
Word	.docx	Document text and structure
Excel	.xlsx	Tabular data
Images	.png, .jpg, .jpeg, .gif, .webp	The image itself (visual analysis)
Plain text	.txt, .md, .csv, .json, .xml	File contents as-is

How it works

When you attach a file to a message:

1. The file is uploaded and stored under your account
2. Content is extracted based on the file type
3. The extracted text is included as context in the conversation
4. The agent can reference the file content when responding

Files live in the workspace you upload them to, under your own private partition — another member of that workspace can’t see your files, and a file uploaded in one workspace isn’t reachable from another. A files://<id> reference resolves within the workspace the conversation belongs to.

Note

Images are sent directly to the model for visual analysis — the agent can describe what it sees, read text in screenshots, interpret charts, and more. This requires a vision-capable model (for example, Claude). Text-based formats are extracted and included as text alongside your message.

Tips

• Be specific about what you want. Attaching a PDF and asking “Summarize the key financial metrics from Q3” gets a more focused answer than “Summarize this.”
• One file at a time works best for detailed analysis. You can attach multiple files, but the agent handles focused requests better.
• Screenshots work great for asking about UI issues, error messages, or visual content.

Feature flag

File context is controlled by the fileContext feature flag, which defaults to true. To disable it:

nimblebrain.json

{
  "features": {
    "fileContext": false
  }
}