Migrate from OpenAI

Why migrate

• One key for many models: GPT, Claude, Gemini, DeepSeek, Kimi, Qwen — same endpoint.
• ~20% cheaper: average price 20% below official, more for high volume. Failed calls aren't billed.
• Low-latency in-region: in-country PoPs + Cloudflare.
• One bill: merged invoice, simpler accounting.
• OEM white-label: we run the infra, you run the brand — own domain + console.

Code changes

The only change is baseURL. Request shape, fields, and responses match OpenAI byte-for-byte.

diff
# Python
  from openai import OpenAI

  client = OpenAI(
-     api_key="sk-openai-xxx",
+     api_key=os.environ["AICODEDOG_KEY"],
+     base_url="https://api.aicodedog.com/v1",
  )

  resp = client.chat.completions.create(
-     model="gpt-4o",
+     model="gpt-5.4",  # or keep gpt-4o — we forward to OpenAI
      messages=[{"role": "user", "content": "Hi"}],
  )

diff
// Node.js
  const client = new OpenAI({
-   apiKey: process.env.OPENAI_API_KEY,
+   apiKey: process.env.AICODEDOG_KEY,
+   baseURL: 'https://api.aicodedog.com/v1',
  });

Migration checklist

1. Create an AICodeDog API key: sign up → Console → API keys.
2. Set env vars: add AICODEDOG_KEY to .env and Vercel / Cloudflare / AWS secrets.
3. One-line baseURL change: add baseURL: "https://api.aicodedog.com/v1" to every client constructor.
4. Run a smoke test: hit one minimal prompt locally, confirm 200.
5. Canary rollout: route 10% of prod traffic for 1–2 days, watch latency / error rate / token usage. Cut over to 100% once happy.
6. Compare metrics: in the dashboard, token usage, error rate, p99 should match or beat your previous numbers.

Model-name compatibility

We support both directions:

• Original OpenAI model names (gpt-4o, gpt-4-turbo, text-embedding-3-small, …) still work — we forward to OpenAI upstream.
• To use models from other vendors, see /models for the slug, e.g.:
  • claude-opus-4-1 ← Anthropic flagship
  • gemini-3-1-pro ← Google flagship
  • DeepSeek-V4-Pro ← best value
  • Kimi-K2.6 ← long context

Edge cases (rare)

• Assistants API / Realtime API: not yet proxied. If you rely on them heavily, keep the official key for that surface and route everything else through AICodeDog.
• Files API (fine-tune uploads): not yet proxied. Fine-tune still goes direct to OpenAI.
• Organization header (OpenAI-Organization): ignored — no effect on results.
• Response id: we re-issue with prefix chatcmpl- so you can audit per call. The upstream OpenAI id is exposed in upstream_id (admin-only).