Core Concepts

A short orientation. Four ideas that show up across every page.

1. Models are namespaced by owner#

Every model ID follows the pattern owner/model-name.

openai/gpt-4o-mini
anthropic/claude-sonnet-4
google/gemini-2.5-pro

The prefix is the company that built the model, not a reseller. openai/gpt-4o always refers to the model OpenAI ships. This keeps naming stable as the ecosystem grows and makes it easy to filter by owner in logs and billing.

2. One API key belongs to one team#

API keys are scoped to a team, not a single user. A key can be used by multiple people or by multiple services. The dashboard shows who created each key and when it was last used.

You can rotate keys at any time and revoke a compromised key with one click. opengateway keeps an audit trail of every change.

See API Keys for the details.

3. Chat fallbacks are a first-class feature#

Chat Completions requests can include a priority list of fallback models. If the first target fails, times out, or is rate-limited, opengateway tries the next one. Fallbacks are handled at the gateway, not in an SDK, so they work the same way from cURL, Python, Rust, or an old Go binary someone left running in production.

{
  "model": "anthropic/claude-sonnet-4",
  "extra": {
    "fallbacks": ["openai/gpt-4o", "google/gemini-2.5-pro"]
  },
  "messages": [...]
}

See Fallbacks for the full behavior.

4. Observability is built in#

Every request through the gateway is logged and costed. There is no middleware to add and no agent to install. You open the dashboard, and the data is already there.

See Observability.

The big picture#

opengateway gives you the Gateway and built-in Observability under the same API key and the same dashboard.