evalops-openai

Drop-in OpenAI SDK shim for routing OpenAI-compatible requests through EvalOps llm-gateway with organization scope, principal attribution, audit, metering, and governance hooks.

This package wraps the official OpenAI SDKs instead of reimplementing OpenAI wire formats, so teams can route model calls through EvalOps with minimal code changes.

Python

pip install evalops-openai

from evalops_openai import OpenAI

client = OpenAI(
    organization_id="org_123",
    principal="user:[email protected]",
)

response = client.chat.completions.create(
    **client.with_provider_ref(
        {
            "model": "gpt-4.1-mini",
            "messages": [{"role": "user", "content": "hello"}],
        }
    )
)

Node

npm install @evalops/openai

import { OpenAI } from "@evalops/openai";

const client = new OpenAI({
  organizationId: "org_123",
  principal: "user:[email protected]",
});

const response = await client.chat.completions.create(
  client.withProviderRef({
    model: "gpt-4.1-mini",
    messages: [{ role: "user", content: "hello" }],
  }),
);

Environment

• EVALOPS_API_KEY or OPENAI_API_KEY: Platform-issued bearer token.
• EVALOPS_ORGANIZATION_ID: organization scope stamped into gateway requests.
• EVALOPS_PRINCIPAL: optional actor string for audit attribution.
• EVALOPS_TRACE_ID: optional trace correlation ID.
• EVALOPS_LLM_GATEWAY_URL: EvalOps LLM gateway base URL.
• EVALOPS_PROVIDER_ENVIRONMENT: defaults to prod.
• EVALOPS_PROVIDER_CREDENTIAL_NAME: optional provider ref credential name.
• EVALOPS_PROVIDER_TEAM_ID: optional provider ref team ID.

Provider References

Use with_provider_ref or withProviderRef when a request should select a specific provider credential. Organizations with default provider routing can omit the helper and keep the vendor SDK call shape.