Developers

One base URL, your existing SDK, and a gateway that treats reliability, latency and honest billing as API features. Everything you need to build is on this page.

Get API key API reference

Start here

The whole migration

Python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.switchgate.ai/v1",  # the only change
    api_key="sg-live-…",
)

r = client.chat.completions.create(
    model="switchgate/auto",
    messages=[{"role": "user", "content": "Bonjour!"}],
)

OpenAI-compatible. Official SDKs in Python, JavaScript, Node.js, Go and plain cURL all work by changing base_url.
Framework-friendly. LangChain, LlamaIndex, the Vercel AI SDK and anything else with an OpenAI provider Just Works - point it at the base URL.
Exact cost on every response via x-sg-cost and in the usage object - streams included.

Resources

Everything, one click deep

Documentation →

Quickstart, governance, caching, fallbacks and the Model Tunnel, written to be read.

API reference →

Every endpoint with parameters, examples, error codes and response headers.

OpenAPI 3.1 spec →

Machine-readable surface for Postman, Bruno, or your codegen pipeline.

Model catalog →

24 models with list-price pass-through, context windows and dedicated pages.

Status →

Live uptime by provider, and the effective success rate after failover.

Changelog →

Every ship in order. Subscribe by RSS.

Benchmarks →

GateBench methodology: task success, effective success, cost per solved task.

Pricing calculator →

Estimate a workload across all 24 models before you route a single token.

llms.txt →

A model-readable map of this site, because your agent reads docs too.

Gateway principles

What we optimize for

Latency honesty: the gateway adds one hop; we publish its overhead on the status page instead of pretending it's zero.
Reliability as a parameter: retries and cross-provider failover are request options (sg_fallbacks), not a sales conversation.
Billing you can audit: list price pass-through, one fee at top-up, exact cost per call, cache hits at literally zero.
Keys that fail safe: prepaid balances, budgets, model locks, IP and country pins, per-key rate limits, instant revocation.

Ship something this afternoon

Free trial credits and the models list is one authenticated GET away.

Get API key Read the docs