Calling OpenAI direct is the simplest possible setup. Adding a gateway buys you a cost dashboard, spend alerts, provider portability, and multi-tenant billing without changing application code. Here is the honest tradeoff.
Sign up in 60 seconds. Top up via Stripe when you are ready.
baseURL and apiKey. Your OpenAI SDK does not change. Updated for what each setup actually delivers as of May 2026.
| Feature | ScopeVeil | OpenAI direct |
|---|---|---|
| OpenAI SDK compatible | Yes | Yes (by definition) |
| Latency OpenAI direct wins on raw latency. The gateway adds tens of milliseconds, which usually does not matter for chat. | Direct call plus a small fixed overhead | Direct call, no extra hop |
| Cost dashboard across all models | Built-in, per-request, with markup breakdown and XLSX export | The OpenAI usage page, per OpenAI key |
| Spend alerts and hard caps | Per-project caps, alerts at thresholds, prepaid credits | Soft and hard limits on the account |
| Multi-provider portability | Switch or split between OpenAI, Anthropic, Google, Mistral, Cohere, Groq, plus 9 more | OpenAI only |
| Multi-tenant billing per project or customer | First-class. Scoped keys, per-org credit pools, exports | One usage view per account |
| Observability for free | Latency, tokens, cost, error rate per route, free up to 500k events/month | You build it (or buy an observability vendor) |
| Provider catalog If you only ever want OpenAI, direct is simpler. The gateway pays off when you might want options. | Multi-provider, curated to first-party APIs | OpenAI models only |
| Privacy posture Both privacy postures are documented. ScopeVeil enforces metadata-only at the SDK schema level. | Operational metadata only; prompt content excluded by schema at ingest | Per the OpenAI data policy |
| Self-hosted option Direct OpenAI is hosted by definition. ScopeVeil offers a self-hosted bundle if compliance requires. | Yes, on Enterprise (full stack in your VPC) | Not applicable (hosted SaaS) |
Highlights mark areas where the gateway changes the picture. OpenAI direct is the right call for simple setups. The gateway pays off when cost visibility, alerts, or provider optionality start to matter.
Every request shows up in the dashboard with model, tokens, latency, and cost. Set alerts when daily spend crosses a threshold. No more opening the OpenAI usage page on the 28th and panicking.
See the pricing calculator →The day OpenAI raises prices, has an outage, or ships a model that is worse than the alternative, you can switch by changing a model string. Application code does not change. No emergency refactor.
Read the security posture →Scoped API keys, per-project credit pools, exports per client. If you build for end customers (agencies, white-label SaaS, internal teams) you stop reimplementing this every quarter.
Quickstart in 60 seconds →If you are calling OpenAI directly with the OpenAI SDK, here is the diff.
import OpenAI from 'openai';
const ai = new OpenAI({
- baseURL: 'https://api.openai.com/v1',
- apiKey: process.env.OPENAI_API_KEY,
+ baseURL: 'https://gateway.scopeveil.com/v1',
+ apiKey: process.env.SCOPEVEIL_KEY,
});
// Same SDK. Same model strings. Same response shape.
const r = await ai.chat.completions.create({
model: 'openai/gpt-4o-mini',
messages: [{ role: 'user', content: 'hello' }],
});That is the whole migration. Even default baseURL setups need only two lines to change. Roll one service first, compare the dashboard to your OpenAI usage page for a week, then roll the rest.
For a single product with one OpenAI account, that is fine. The gateway starts paying off when you want a cost dashboard with alerts, the ability to switch providers, or per-project billing. If you do not need those today, skip it.
A small fixed overhead per request, typically in the tens of milliseconds. For streaming we forward bytes as they arrive, so time-to-first-token tracks OpenAI closely. For most chat and agent workloads the overhead is well below the network jitter of the upstream call itself.
Each request reserves an estimate, hits OpenAI, then we charge the actual upstream cost plus your tier markup (5 to 15 percent). The dashboard splits the two so the markup is never hidden. For many teams the analytics save more than the markup costs.
Yes. Many teams start by routing one service through ScopeVeil and keep the OpenAI key for the rest. You can also use the observability SDK without the gateway if you only want the dashboard.
The gateway is a single dependency, which is a fair concern. Status is published, error rates are graphed, and the SDK can fall back to direct OpenAI on configurable error classes. For regulated workloads the self-hosted bundle removes this concern entirely.
For the hosted gateway, request and response bodies are read in-memory for token counting and discarded. We persist metadata only (model, tokens, latency, cost, your tags). For the self-hosted bundle, nothing leaves your VPC.
Top up via Stripe and send your first request in under a minute.