Migrate to OpenDunes in One Prompt

If your app already calls the OpenAI API, switching to OpenDunes is two changes: the base URL and the key. Every request shape, SDK method, and streaming behavior stays identical.

The two-line change#

Wherever you construct your OpenAI client, update baseURL and apiKey:

# Before
from openai import OpenAI
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# After
from openai import OpenAI
client = OpenAI(
    base_url="https://opendunes.com/v1",
    api_key=os.environ["OPENDUNES_API_KEY"],
)

That's it. All your existing chat.completions.create(...) calls work without further modification.

Update your model slugs#

OpenDunes uses provider/model slugs instead of bare model names. Swap them out:

Browse the full catalog at /models — every listing shows the exact slug to use.

Update your environment variables#

Add the new key and (optionally) keep the old one during a parallel-run period:

# .env (or your CI/CD secrets)
OPENDUNES_API_KEY=sk-0a1b2c3d4e5f...
# OPENAI_API_KEY=sk-...  ← keep until traffic is fully migrated

Get your OpenDunes key from Dashboard → Keys. If you haven't created a custom key yet, the default key is ready to use immediately after signup.

What stays the same#

Everything in chat/completions that you use today transfers over directly:

• messages array with system, user, assistant roles
• stream: true with the same SSE delta format
• max_tokens, temperature, top_p
• Multi-part message content for vision inputs
• Error HTTP codes (401, 429, 5xx) and retry logic

Billing in DA#

OpenDunes bills in Algerian Dinars (DA). Before your first request, deposit DA with a CIB or Edahabia card from Dashboard → Billing. Pricing is shown per million tokens in DA on the models page.

The response header X-Balance-Available tells you your remaining balance after every call, so you can build low-balance warnings without polling.

Check for unsupported features#

Almost everything transfers. Tool/function calling, JSON mode, structured outputs (response_format), and embeddings (/v1/embeddings) are all supported today. The Responses API is carried too: /v1/responses takes the same body, and its errors come back in the Responses shape — see Responses.

Validation is forgiving: unknown fields in the request body are simply ignored, and an invalid field value returns a 422 validation error naming the field to fix.

Everything else — chat completions, streaming, vision inputs, multi-turn history — works today. There is no cap on the number of requests; a key's limit is a spending cap in DA.

Verify the migration#

A quick sanity check after you've updated the client:

curl https://opendunes.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-5",
    "messages": [{ "role": "user", "content": "Reply with OK." }]
  }'

A 200 with a valid choice means the migration is complete.

Next steps#

• Streaming — SSE format details and error handling.
• Models — Full catalog with DA pricing.
• Error reference — Error codes and retry strategy.