oriora 0.1.0

Python client for Oriora — model selection (decision-only) + OpenAI-compatible routing (Managed API BYOK).

Oriora Python SDK

Thin Python client for Oriora — Managed API BYOK.

Two ways to use it:

• model_select() — decision only. Oriora tells you the best model for a task; you run the call yourself with your own vendor key. Oriora never sees your key, prompt, or output. Flat $0.001 per decision.
• chat() — OpenAI-compatible. Oriora routes and executes the call server-side using the BYOK vendor key you configured, and returns an OpenAI-shaped response.

Install

pip install oriora

Authenticate

Generate an sk_oriora_ API key in your Oriora account → Settings. Pass it directly or set ORIORA_API_KEY.

from oriora import Oriora

client = Oriora(api_key="sk_oriora_...")   # or: Oriora()  with ORIORA_API_KEY set

model_select() — decision only (you run the call)

rec = client.model_select(task_type="coding")
# {'model': 'anthropic/claude-sonnet-4.6', 'alternatives': [...], 'task_type': 'coding'}

# Then call that model yourself, with your own vendor key:
import anthropic
anthropic.Anthropic().messages.create(model="claude-sonnet-4.6", messages=[...])

Restrict the recommendation to your own candidate list (your quality order is kept):

client.model_select(task_type="coding", models=["openai/gpt-5", "anthropic/claude-sonnet-4.6"])

Discover the valid task types:

client.task_types()
# ['agentic', 'coding', 'general', 'math', 'reasoning', ...]

chat() — OpenAI-compatible (we run the call via your BYOK key)

First, in your Oriora account, store your vendor key and enable BYOK for an app label (e.g. my-app). Then:

resp = client.chat(
    app="my-app",                # the BYOK app label you enabled
    messages=[{"role": "user", "content": "Explain merge sort in one line."}],
    # model defaults to "oriora-auto" — Oriora picks the best model for your prompt
)
print(resp["choices"][0]["message"]["content"])

The response is an OpenAI-shaped ChatCompletion dict (id, choices, usage, …), so existing OpenAI-style code works with minimal changes.

Note: usage token counts on chat() are currently estimated. Specific-model pinning and streaming are not yet supported (use model="oriora-auto").

Errors

Non-2xx responses raise OrioraError (with .status_code and .message).

from oriora import OrioraError
try:
    client.model_select(task_type="coding")
except OrioraError as e:
    print(e.status_code, e.message)

Pricing

• model_select() — flat $0.001 / decision.
• chat() — Managed API BYOK orchestration fee on the call (your vendor bills you directly for the AI usage on your own key).