Multimodal input

Images, audio & documents as input

Send an array of content parts instead of a string, and a model that accepts that modality will read them. Both the OpenAI spelling and the Responses spelling are understood on /v1/chat/completions:

Part	Carries	Source
`image_url` / `input_image`	An image	`{"url": "https://…"}` or a `data:image/…;base64,…` URL
`input_audio`	An audio clip	`{"data": "<base64>", "format": "wav"}`
`file` / `input_file`	A PDF or other document	`{"filename": "…", "file_data": "data:application/pdf;base64,…"}`

import base64, os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["OXYY_API_KEY"],
    base_url="https://api.oxyy.ai/v1"
)

# A public https URL, or a data URL you build yourself.
with open("chart.png", "rb") as f:
    b64 = base64.b64encode(f.read()).decode()

response = client.chat.completions.create(
    model="gemini-3-flash",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "What does this chart show?"},
            {"type": "image_url", "image_url": {
                "url": f"data:image/png;base64,{b64}",
                "detail": "auto",
            }},
        ],
    }],
)
print(response.choices[0].message.content)

curl https://api.oxyy.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OXYY_API_KEY" \
  -d '{
    "model": "gemini-3-flash",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "Describe this image."},
        {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}}
      ]
    }]
  }'

# A PDF or other document rides in a `file` content part.
import base64, os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["OXYY_API_KEY"],
    base_url="https://api.oxyy.ai/v1"
)

with open("contract.pdf", "rb") as f:
    b64 = base64.b64encode(f.read()).decode()

response = client.chat.completions.create(
    model="gemini-3-flash",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "Summarise the termination clause."},
            {"type": "file", "file": {
                "filename": "contract.pdf",
                "file_data": f"data:application/pdf;base64,{b64}",
            }},
        ],
    }],
)

A model only reads what it declares. Each model lists its inputModalities in GET /v1/models. Attaching an image to a text-only model is refused rather than quietly dropped, because a silently ignored attachment produces a confident answer about nothing. Per-file caps are 20 MB for images, 50 MB for audio and 100 MB for video and documents, and an operator can set a lower cap per model.

URLs are fetched by us, not by the provider. A URL you supply is downloaded, size-checked and inlined before the provider call, so private and loopback addresses are refused. If a URL is unreachable you get a 400 that says so, not a confusing provider error.