OpenGateway documentation

OpenGateway is the open-source AI gateway for the best open-weight models. Point any OpenAI- or Anthropic-compatible client at one base URL and call Kimi, MiniMax, Qwen, DeepSeek, GLM and GPT-OSS — routed across live inference providers at the edge.

Open weights only

These docs cover the public /oss lane — 100% open-weight, self-hostable models. No proprietary frontier lock-in in the default path.

One base-URL swap

OpenGateway runs as a Cloudflare Worker at the edge and speaks both the OpenAI and Anthropic wire protocols on the /oss lane. Bring your existing coding agent or SDK, change the base URL, and it just works — no code changes, no forks.

curl · /oss/v1/chat/completions

# OpenAI-compatible — just swap the base URL
curl https://api.opengateway.one/oss/v1/chat/completions \
  -H "Authorization: Bearer $OPENGATEWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.2",
    "messages": [{"role": "user", "content": "Refactor this function"}]
  }'

Base URLs

Everything public lives under https://api.opengateway.one/oss.

Surface	URL	Use for
OSS lane root	`https://api.opengateway.one/oss`	Anthropic SDK, Claude Code (CLI/IDE/Desktop)
OpenAI-compatible	`https://api.opengateway.one/oss/v1`	OpenAI SDK, Codex, Cursor
Model discovery	`https://api.opengateway.one/oss/v1/models`	Catalog + agent pickers
Live status	`https://api.opengateway.one/status/api`	Machine-readable health

The /v1 gotcha

The OpenAI base URL includes /v1. The Anthropic base URL excludes it — the SDK appends its own /v1/messages. Both land on the gateway's /v1/* routes.

Why OpenGateway

One API for any open model — drop-in compatible with the OpenAI and Anthropic SDKs.
Higher availability — every model is hosted by multiple providers; the gateway routes to a healthy one and fails over automatically.
Price & performance — pick a routing policy per request: :preferred, :fastest or :cheapest.
100% open weights — every model is open-weight and self-hostable, so you keep portability and control of your data.
Measured telemetry — real provider data on status.opengateway.one, never vanity metrics.

Start here

Quickstart

Swap your base URL and send your first /oss request in under five minutes.

Lanes & routing

How the /oss catalog and :fastest/:cheapest/:preferred policies work.

API reference

Chat completions, messages, models and embeddings — with copy-paste examples.

Client setup

Copy-paste config for Claude Code (CLI, IDE, Desktop), Cursor and Codex.