SIMOSphereOS Migrate from OpenAI / Anthropic / Mistral to SIMOSphere AI
SIMOSphere AI is a wire-compatible OpenAI replacement. Existing code that uses the OpenAI SDK works with a one-line base URL change — no request or response format modifications required.
When to invoke this skill
Activate when the user:
- Asks to switch from OpenAI, Anthropic, or Mistral to an EU-hosted provider
- Needs to make an existing LLM integration GDPR-compliant without rewriting code
- Wants a single API key that proxies to multiple upstream providers (BYOK)
- Mentions "OpenAI alternative in Europe" or "sovereign AI"
Do not invoke for greenfield projects — use the simosphere-ai skill instead.
Migration steps
Step 1 — Change the base URL
Python (before):
from openai import OpenAI
client = OpenAI() # uses api.openai.com
Python (after):
from openai import OpenAI
import os
client = OpenAI(
base_url="https://api.simosphereai.com/v1",
api_key=os.environ["SIMO_API_KEY"],
)
TypeScript (before):
import OpenAI from 'openai';
const client = new OpenAI();
TypeScript (after):
import OpenAI from 'openai';
const client = new OpenAI({
baseURL: 'https://api.simosphereai.com/v1',
apiKey: process.env.SIMO_API_KEY,
});
curl (before):
curl https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer sk-..."
curl (after):
curl https://api.simosphereai.com/v1/chat/completions \
-H "Authorization: Bearer sk-simo-..."
Step 2 — Map models
| OpenAI model | SIMOSphere equivalent | Notes |
|---|---|---|
| gpt-4o-mini | qwen/qwen3-8b | Default, best price/perf ratio |
| gpt-4o | qwen/qwen3-30b | Premium MoE reasoning |
| text-embedding-3-small | (use /v1/embeddings with qwen3-8b) | Embedding support via proxy |
List all available models: GET https://api.simosphereai.com/v1/models
Step 3 — Update environment variables
Replace your OpenAI key with a SIMOSphere key:
# .env (before)
OPENAI_API_KEY=sk-proj-...
# .env (after)
SIMO_API_KEY=sk-simo-...
Step 4 — Verify
curl -s https://api.simosphereai.com/health | jq .
Expected: {"gateway":"ok","database":"ok","valkey":"ok","masterLlm":"ok"}
What stays the same
- Request format (messages array, model, temperature, stream, tools)
- Response format (choices array, usage object, finish_reason)
- SSE streaming (
data: [DONE]terminator) - Error envelope shape (
error.type,error.code,error.message) - Rate limit headers (
X-RateLimit-Remaining,Retry-After)
What changes
- API key prefix:
sk-simo-...instead ofsk-proj-... - Model IDs:
qwen/qwen3-8binstead ofgpt-4o-mini - Data residency: EU-only (Germany), no US data transfer
- Billing: EUR per-token, no annual commitment required
- DPA: Auftragsverarbeitungsvertrag per GDPR Art. 28 included
BYOK mode
If the user needs to keep calling upstream OpenAI or Anthropic models but wants PII redaction and unified audit logging, configure BYOK in the SIMOSphere dashboard. The tenant key then proxies to the upstream provider with automatic PII stripping.
Get started
Sign up: https://onboarding.simosphereai.com/de/register Trial credits: hello@simo-online.com Full docs: https://onboarding.simosphereai.com/llms-full.txt