Migrate from OpenAI

SovereignEG's API is OpenAI-compatible for chat completions, embeddings, image generation and model listing. For those, migration means changing your API key and base URL and choosing a model from the catalog. Other OpenAI APIs (Responses, Assistants, Files, Batches, Audio, fine-tuning, image edits and the legacy /v1/completions) are not available.

Step 1: Change the base URL

Python

# Before (OpenAI)
from openai import OpenAI
client = OpenAI(api_key="sk-openai-...")
 
# After (SovereignEG) — one line changed
from openai import OpenAI
client = OpenAI(
    api_key="sk-...",
    base_url="https://backend.sovereigneg.com/v1"
)

Node.js

// Before (OpenAI)
const client = new OpenAI({ apiKey: 'sk-openai-...' });
 
// After (SovereignEG) — one line added
const client = new OpenAI({
  apiKey: 'sk-...',
  baseURL: 'https://backend.sovereigneg.com/v1',
});

curl

# Before
curl https://api.openai.com/v1/chat/completions ...
 
# After — just change the URL
curl https://backend.sovereigneg.com/v1/chat/completions ...

Step 2: Map your models

Some OpenAI models are listed under their usual OpenAI ids — check the Model Library or GET /v1/models for the exact id (an id that isn't listed returns 400 model_not_found). For anything not listed, find a live equivalent:

Step 3: Check the differences

Chat completions, streaming (SSE), tool calling, system messages, multi-turn conversations, temperature / top_p / max_tokens and stop sequences use OpenAI's request and response format. Watch for these differences:

• Streaming: token usage is always reported, so every stream ends with a chunk whose choices list is empty. Check choices before reading choices[0].
• Legacy function calling: the deprecated top-level functions / function_call parameters are not supported; use tools / tool_choice.
• Sampling on Claude Opus 4.7 and later: these Claude models accept only their default sampling, so temperature and top_p are not sent to them while Anthropic serves the request. The request still succeeds, and the X-Dropped-Params response header names what was left out (see Chat Completions).
• Moderation: /v1/moderations is a basic keyword check, not an AI moderation model. Its model field reads sovereigneg-basic-keyword-check. Keep your own moderation step for production content.
• Errors: the JSON error object has the same shape (plus request_id), but some status codes differ, so check how your code branches on them: running out of credit or hitting a spend cap returns 402 (type insufficient_quota; OpenAI uses 429), an unknown model returns 400 model_not_found (OpenAI uses 404), and a body that fails schema validation returns 422 invalid_request_body (OpenAI uses 400). A 429 retried inside the 60-second window is counted and rejected again, so stop sending for at least the Retry-After time instead of relying on the SDK's automatic retries. See Errors.

Environment variable approach

The cleanest migration uses environment variables:

import os
from openai import OpenAI
 
client = OpenAI(
    api_key=os.environ["LLM_API_KEY"],
    base_url=os.environ.get("LLM_BASE_URL", "https://api.openai.com/v1")
)

# Switch providers by changing env vars — no code change
LLM_API_KEY=sk-...
LLM_BASE_URL=https://backend.sovereigneg.com/v1

What's different

LangChain migration

If you're using LangChain, it's the same one-line change:

from langchain_openai import ChatOpenAI
 
llm = ChatOpenAI(
    model="gpt-oss-20b",
    api_key="sk-...",
    base_url="https://backend.sovereigneg.com/v1"
)

LangChain chains and agents that use ChatOpenAI with tool calling (bind_tools) work with this change, on models that support tool calling. Not supported: the Responses API (use_responses_api=True or OpenAI built-in tools; the endpoint returns 404), the legacy functions parameter used by create_openai_functions_agent and bind_functions (it is ignored, so the model never sees those functions), and wrappers for OpenAI's Assistants, audio and files APIs.