Send an image or PDF attachment

Upload a file directly to MinIO via a presigned URL, then reference it in a chat message.

The goal

A user message that includes an image or PDF; the agent processes it as a multimodal input on a vision-capable model, or text-falls-back on a non-vision model.

Steps

1. Get a presigned URL.

   const presigned = await platos.attachments.presigned({
     filename: "invoice.pdf",
     mimeType: "application/pdf",
     conversationId,
   });
   

2. Upload directly to MinIO.

   await fetch(presigned.url, {
     method: "PUT",
     body: fileBlob,
     headers: { "Content-Type": "application/pdf" },
   });
   

   Bytes never touch the agent runtime.

3. Send the message.

   await platos.threads.update({
     threadId,
     messages: [{
       role: "user",
       content: "Summarise this invoice",
       attachments: [presigned.attachmentId],
     }],
   });
   

4. Stream the response.

   The agent receives the attachment in its turn assembly; the multimodal adapter routes it to the model.

Verify

• Postman mode shows the attachment id in the assembled message.
• The trace shows a multimodal.adapt span with mimeType: application/pdf.
• The reply references the attachment's content.

Vision routing

When the agent's model has vision: true, image bytes (or signed URLs) are inlined. PDFs are converted to page images plus extracted text. Non-vision models receive a text fallback ("the user attached an image: ") plus a tool call to platos-code-runner if image processing is needed.

Limits

• Default per-file cap: 50 MB.
• Default mime types: image/png, image/jpeg, image/webp, image/gif, application/pdf, text/csv, text/plain, application/json.
• Retention: project default (configurable, see Attachments and files).

Next steps

• Filter PII before the attachment text reaches the model.
• See platos-code-runner for sandboxed PDF extraction.