ngrok AI Gateway
Connect self-hosted models and leading model providers through one managed gateway with built-in observability and access control.
About ngrok AI Gateway
ngrok AI Gateway brings self-hosted models and leading model providers into one managed gateway for unified access. An application only needs to change its baseURL to gateway.ngrok.ai and replace its API key to connect; locally deployed LLMs can also be reached directly by the gateway over a private connection, without dealing with public IPs or inbound ports. You can centrally manage OpenAI, Anthropic, or custom provider keys, issue separate access keys to each application or developer, and define their available providers and models. The gateway aggregates tokens, latency, and errors from all calls; when a model or key slows down or fails, requests are rerouted to healthy fallbacks you define.
Best for
Best use cases
Change the baseURL to connect to the gateway and unify calls to cloud providers and self-hosted local models
Issue a separate access key to each application or developer and limit its available providers and models
Manage OpenAI, Anthropic, or custom provider API keys centrally
View tokens, latency, and errors across all calls and reroute to healthy fallbacks when a model or key fails
Features
Unified managed gateway
Bring self-hosted models and leading model providers into one managed gateway with built-in observability and access control.
Local model access
The gateway can route to reachable local LLMs over a private connection without dealing with public IPs or inbound ports.
Keys and access control
Centrally manage OpenAI, Anthropic, or custom provider keys; give each application or developer a separate access key and limit its available providers and models.
Observability and failover
Aggregate tokens, latency, and errors across all calls; when a model or key slows down or fails, route requests to healthy fallbacks you define.
Pros
- Connect by changing only the baseURL and API key
- Connect local LLMs over a private connection without dealing with public IPs or inbound ports
- Manage keys centrally and limit the providers and models each application or developer can call
- Aggregate tokens, latency, and errors across all calls, and reroute to healthy fallbacks during failures
Pricing Plans
Pay as you go
- Plus the inference cost itself
- Inference is passed through at cost with an ngrok key
- Your own provider key is billed directly by the provider
* Pricing information is for reference only. Please refer to the latest announcement on ngrok AI Gateway's official website.