Multimodal

Nodal supports multimodal inputs where the selected model and route support them.

Chat App Attachments

The first-party chat app can upload files to the active workspace and attach file IDs to chat sends. Files are scanned asynchronously and remain workspace-scoped.

When the selected model supports vision, image attachments can be sent to the model after Nodal resolves file references to signed object URLs.

Gateway Image Parts

/v1/chat/completions accepts OpenAI-style content blocks, including image URL parts:

{
  "model": "Vision",
  "messages": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "Describe this image." },
        { "type": "image_url", "image_url": { "url": "https://example.com/image.png" } }
      ]
    }
  ]
}

Configured file://<uuid> image references are resolved to workspace-scoped signed URLs before provider dispatch.

Error Behavior

If a model does not support vision, Nodal returns an OpenAI-shaped error with code model_does_not_support_vision.

If file references are unavailable, not ready, or not found, the gateway returns file_references_not_supported, file_not_ready, or file_not_found.

Knowledge Files

Knowledge collections can ingest text and supported document files. Documents are parsed, chunked, embedded, and made searchable inside the workspace.