Inference API (OpenAI-compatible)

One endpoint speaks the protocol every AI library already knows. If your code works with OpenAI, it works here — change base_url, keep everything else.

from openai import OpenAI

client = OpenAI(base_url="https://platform.senaiy.ai/v1", api_key="sk-sf-...")

What is supported

Capability	Endpoint	Status
Chat completions	`POST /v1/chat/completions`	✅ verified in production
Streaming (SSE)	same, `"stream": true`	✅ verified
Vision (image input)	same, `image_url` content on a vision model	✅ verified — see below
Embeddings	`POST /v1/embeddings`	✅ verified
Model listing	`GET /v1/models`	✅ verified — scoped to your key
Speech-to-text	`POST /v1/audio/transcriptions`	✅ verified — availability follows your key's allowlist
Image generation / Assistants / Files / Fine-tuning	—	❌ not offered — use Agents & Knowledge instead

One interface, many providers

The catalog spans OpenAI (gpt-4o, gpt-4o-mini), Anthropic (claude-sonnet, claude-haiku), Google (gemini-flash), regional and local engines — all through the same API shape, all rated in USD against the same wallet. Your key's allowlist decides which of them you see and may call.

for model in client.models.list().data:
    print(model.id)

Streaming

stream = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "اكتب فقرة عن الرياض"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

Vision (image input)

Vision-capable models (gpt-4o, gpt-4o-mini, claude-sonnet, claude-haiku, gemini-flash, …) accept images inline using the standard image_url content part — pass a public URL or a data: URI. Check a model's supports_vision flag in the portal's Developer → Models directory before relying on it.

resp = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": [
        {"type": "text", "text": "What is in this image?"},
        {"type": "image_url", "image_url": {"url": "https://…/photo.png"}},
    ]}],
)

Speech-to-text

resp = client.audio.transcriptions.create(
    model="whisper-large-v3",
    file=open("meeting.mp3", "rb"),
)
print(resp.text)

Which speech models you can call is decided by your key's allowlist, same as any other model.

Residency

Organizations under a local_only residency policy may only call local-badged engines; a cloud model returns residency_policy_violation. That refusal is the feature: data governed to stay in-country stays in-country.

Structured output

Pass response_format with a JSON schema on models that support it and validate the result server-side — or better, put the schema on an agent or workflow node, where the platform validates and retries for you.