Move to llmdeal in 5 minutes, without rewriting your app.

Same OpenAI-compatible API. Same client libraries. Just a new base URL and key. Pick where you're coming from.

Migrating from Cursor's bundled API

Cursor's $20/mo Pro plan bundles a fixed slice of model compute. If you're hitting the cap or you want to control which models your code touches, point Cursor at your own llmdeal endpoint.

What's different

Pricing. Cursor Pro is $20/mo with bundled (rate-limited) compute. llmdeal Fleet is $649/mo flat for a far larger envelope, or Burst Day is $129 for one day of heavy use. No per-token meter.
KYC. Cursor requires a card on file. llmdeal accepts BTC/XMR/LTC; no name, no card.
Residency. Cursor routes through whichever upstream answers fastest. llmdeal lets you pin EU-only or US-only routing per-key, which matters if your code or prompts touch GDPR-relevant data.

The actual swap

Open Cursor → Settings → Models → OpenAI API Key (toggle "Override OpenAI Base URL"). Or edit ~/.cursor/settings.json directly:

~/.cursor/settings.json

  {
    "cursor.aiProvider": "openai",
-   "openai.apiKey": "sk-proj-...",
-   "openai.baseUrl": "https://api.openai.com/v1"
+   "openai.apiKey": "lld_xxxxxxxxxxxxxxxx",
+   "openai.baseUrl": "https://api.llmdeal.me/v1"
  }

Restart Cursor. The composer and tab-complete will now route through llmdeal.

Test it

shell

# Replace lld_... with your key
curl https://api.llmdeal.me/v1/chat/completions \
  -H "Authorization: Bearer lld_xxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{"model":"smart-route","messages":[{"role":"user","content":"hello"}]}'

A 200 with a JSON body confirms the key is live. If you see 401, double-check the key prefix is lld_, not sk-.

Things to watch out for

Streaming responses behave identically; no client changes needed.
Model name remapping: Cursor sends gpt-4 or gpt-4o by default. Those route through smart-route on our side. If you bought the Frontier Credits add-on, pin to claude-sonnet-4-6-frontier for parity with Cursor's stronger model picks.
Rate limits start lower than OpenAI's. Hobbyist is 5 rpm; Fleet is 200 rpm. Check /pricing for your tier.
Tool/function calling: identical schema. Cursor's "agent mode" works as-is.
Cursor caches some completions client-side. Clear the IDE cache if you see stale model responses after switching.

Saving math

A heavy Cursor user pushing past the Pro cap typically pays $20/mo for Pro + $40/mo overage credits = $60/mo. Switching to llmdeal Burst Day ($129) covers ~5 of your heaviest days without overage, or Fleet ($649) covers an entire team. For a solo dev with bursty days, Burst Day at $129 amortises to ~$25/mo if you use it once a fortnight. Numbers reflect llmdeal flat-rate SKUs from /pricing; Cursor pricing varies by overage rate.

Migrating from OpenRouter

OpenRouter is the closest neighbour. Same OpenAI-compatible schema, same multi-model routing concept. The swap is one env var.

What's different

Pricing. OpenRouter is per-token with a small markup. llmdeal is flat-rate: Fleet $649/mo, Fleet Plus $1,199/mo, Burst Day $129. No surprise bills.
KYC. OpenRouter accepts crypto top-ups but caps unverified accounts at $50. llmdeal has no cap and never collects ID.
Residency. OpenRouter routes through whoever's cheapest; you can't pin EU-only. llmdeal lets you set X-Region: eu or buy an EU-only key.

The actual swap

If your code reads OPENAI_API_BASE and OPENAI_API_KEY from env (or the OpenRouter equivalents), edit your .env:

.env

- OPENAI_API_BASE=https://openrouter.ai/api/v1
- OPENAI_API_KEY=sk-or-v1-xxxxxxxxxxxxxxxxxxxxxxxx
+ OPENAI_API_BASE=https://api.llmdeal.me/v1
+ OPENAI_API_KEY=lld_xxxxxxxxxxxxxxxx
  
# If your library reads OPENROUTER_API_KEY specifically,
# also set it to the same lld_ value, or rename to OPENAI_API_KEY.

No SDK changes. The official OpenAI SDK, LangChain, LlamaIndex, Aider, and Continue.dev all pick up the new base URL on next launch.

Test it

shell

curl https://api.llmdeal.me/v1/models \
  -H "Authorization: Bearer $OPENAI_API_KEY"

You should see a JSON list including smart-route, claude-sonnet-4-6-frontier, gpt-5.5-frontier, and the open-weight aliases. Full catalog at /models.html.

Things to watch out for

Streaming responses behave identically; no client changes needed.
Model name remapping: OpenRouter uses vendor/model slugs (e.g. anthropic/claude-3.5-sonnet). On llmdeal those become flat aliases: claude-sonnet-4-6-frontier, smart-route, etc. See /models.html for the full mapping.
Rate limits start lower. Hobbyist is 5 rpm. If you were hammering OpenRouter at 60+ rpm, you'll need Fleet or higher.
Caching headers are passed through. If you used prompt_cache_key on OpenRouter for Anthropic models, it still works on our Sonnet and GPT-5.5 aliases.
Tool/function calling: identical schema. Tested with our Claude, GPT, and Gemini frontier aliases.
Custom HTTP headers like HTTP-Referer and X-Title that OpenRouter uses for app attribution are ignored on llmdeal; harmless to leave in place.

Saving math

A working example: 50k requests/month at 800 prompt + 200 completion tokens averages roughly $145/mo on OpenRouter routing to Sonnet 3.5. On llmdeal Fleet that traffic sits well inside the $649/mo envelope alongside other workloads. For a single project that volume, Burst Day ($129) for one heavy day per week amortises to ~$97/mo, saving ~$48/mo or 33%. Per-token costs vary by model tier; run your own numbers at /calc.

Migrating from OpenAI direct

If you're calling api.openai.com directly, llmdeal is a drop-in replacement at the SDK level. Same client library, same method signatures, new base_url and key.

What's different

Pricing. OpenAI is per-token: GPT-4o roughly $5 input / $15 output per 1M tokens. llmdeal Fleet is $649/mo flat covering equivalent volumes; Burst Day is $129 for a 24-hour heavy session.
KYC. OpenAI requires a card and identity verification for higher-tier rate limits. llmdeal does neither.
Residency. OpenAI's API is US-hosted by default (EU residency requires a separate enterprise contract). llmdeal lets you pick EU or US per-key from sign-up.

The actual swap

Two lines in your client setup. Python:

Python (openai >= 1.0)

  from openai import OpenAI
  
- client = OpenAI(api_key="sk-proj-...")
+ client = OpenAI(
+     api_key="lld_xxxxxxxxxxxxxxxx",
+     base_url="https://api.llmdeal.me/v1",
+ )
  
  resp = client.chat.completions.create(
-     model="gpt-4o",
+     model="smart-route",
      messages=[{"role": "user", "content": "hello"}],
  )

Node.js:

Node.js (openai >= 4)

  import OpenAI from "openai";
  
- const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });
+ const openai = new OpenAI({
+   apiKey: process.env.LLD_KEY,
+   baseURL: "https://api.llmdeal.me/v1",
+ });

Test it

shell

curl https://api.llmdeal.me/v1/chat/completions \
  -H "Authorization: Bearer lld_xxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "smart-route",
    "messages": [{"role":"user","content":"say hi"}]
  }'

Things to watch out for

Streaming responses behave identically; no client changes needed.
Model name remapping: we expose smart-route as the default. Map existing gpt-4 / gpt-4o calls to smart-route, or pin to claude-sonnet-4-6-frontier or gpt-5.5-frontier if you bought Frontier Credits.
Rate limits start lower than OpenAI Tier 4. Hobbyist tier is 5 rpm; Fleet is 200 rpm. Plan accordingly for batch jobs.
Prompt caching: if you used OpenAI's automatic prompt caching (the 50% discount on repeated prefixes), the same prompt_cache_key hint works on our Sonnet and GPT-5.5 frontier aliases.
Tool/function calling, JSON mode, structured outputs, and vision inputs all pass through unchanged on the frontier aliases. Open-weight aliases support tools but not vision; check /models.html for per-model capability flags.
The Assistants API and Realtime API are not implemented. Use Chat Completions with streaming instead.

Saving math

A workload running 50k requests/month with 800-token prompts and 200-token completions on gpt-4o costs roughly $350/mo on OpenAI direct. The same volume routed through llmdeal Fleet ($649/mo) covers it with headroom for a second project; or Burst Day ($129) covers one 24-hour heavy session per fortnight at ~$258/mo equivalent, saving ~$92/mo or 26% for solo workloads, more once you add a teammate. OpenAI per-token rates change; verify against the latest openai.com/pricing when sizing.

Ready to switch?

Buy a key in 60 seconds. No card, no signup form, no email-confirmation loop.

Fleet · $649/mo Burst Day · $129 $1 demo

Migration FAQ

Will my existing API key still work?

No. sk- and sk-or- keys from OpenAI and OpenRouter don't authenticate against llmdeal. You need an lld_ key. Try the $1 demo first if you want to sanity-check before committing.

What about rate limits during a migration?

Your existing provider keys keep working until you swap them. There's no atomic cutover required. Run a few requests through llmdeal, confirm behaviour, then move the rest of your traffic over.

Can I run both in parallel for A/B testing?

Yes, and we recommend it. Most teams send 5% of traffic to llmdeal for a week before switching the remainder. Use two client instances with different base URLs.

What if my model isn't in your catalog?

Check /models.html for the current list. If a model is missing and you'd actually pay to use it, DM the operator via Discord. We add models based on demand.

What's the refund policy?

Credits never expire. Subscription SKUs (Fleet, Fleet Plus) cancel by not renewing; the active month isn't refunded. Burst Day is non-refundable but consumable any time.

Will it slow down my app?

Latency overhead is ~5 to 15 ms on top of the upstream model latency. That's Cloudflare tunnel plus our proxy hop. The model itself responds at the same speed as if you called the vendor directly.

What about logging and observability?

We log request metadata (timestamp, model, token counts) for billing. We do not log prompt or completion content by default. EU-residency keys keep all metadata in EU storage. See /privacy for the full data policy.