refutics 0.1.0

Drop-in token optimization for the OpenAI & Anthropic SDKs — change one import, cut your token bill.

Refutics — Python SDK

Drop-in token optimization. Change one import, cut your token bill. Streaming, tools, and vision keep working — these are thin wrappers over the official SDKs.

Install

pip install refutics openai          # for OpenAI / OpenAI-compatible / Azure
pip install refutics anthropic       # for Anthropic

Use — one import change

# before:  from openai import OpenAI
from refutics import OpenAI

client = OpenAI()                                   # reads REFUTICS_TOKEN from env
resp = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "hello"}],
)

Set your token once:

export REFUTICS_TOKEN=dtras_live_…

Any provider — one token

client = OpenAI(provider="groq")        # or mistral, deepseek, together, xai, fireworks, gemini…

Anthropic

from refutics import Anthropic
client = Anthropic()                                # reads REFUTICS_TOKEN
client.messages.create(model="claude-3-5-sonnet-latest", max_tokens=256, messages=[...])

Azure OpenAI

from refutics import AzureOpenAI
client = AzureOpenAI(
    azure_resource="https://my-company.openai.azure.com",
    api_version="2024-02-01",
)
client.chat.completions.create(model="<your-deployment>", messages=[...])

Provider key

• Stored (recommended): add your provider key once in the Refutics dashboard — pass nothing.
• BYOK: OpenAI(provider_key="sk-…") — never stored.

Optional: auto model-routing (save more)

client = OpenAI(route=True)   # downgrades clearly-easy prompts to a cheaper model

That's it — same SDK, same answers, smaller bill.