About ngrok AI Gateway
ngrok AI Gateway is a hosted gateway that routes requests to AI models across public providers, custom endpoints, and self-hosted models through a single URL and API key. It launched this week on Product Hunt and is built by the team behind ngrok's tunnel infrastructure. The gateway connects private models through ngrok's network so they sit alongside hosted providers without exposing them to the public internet.
Review
ngrok AI Gateway addresses a problem that shows up quickly when you start building with multiple AI providers: you end up juggling accounts, SDKs, dashboards, and fallback logic. The product consolidates routing, observability, and access control into one hosted service. It launched this week, so some features are still in active development while others are already available.
Key Features
- One URL (gateway.ngrok.ai) and one access key route requests to OpenAI, Anthropic, custom endpoints, and self-hosted models
- Self-hosted models connect through ngrok's network, so they don't require opening inbound ports or exposing IPs
- Fallback definitions let you list models in priority order; when one fails or times out, the gateway retries the request with the next candidate
- Bring your own provider keys or use ngrok-managed credits to make requests against providers like OpenAI, Anthropic, and z.ai
- Usage tracking shows tokens, latency, errors, models, providers, and estimated cost in one dashboard
- Access control works per application or developer: separate keys determine which providers and models each caller can use
Pricing and Value
The Product Hunt listing does not include pricing details. The product description mentions using ngrok.ai credits to make requests against providers without creating your own accounts, but no credit pricing is published. You can also bring existing provider keys, which means you'd pay those providers directly and use the gateway for routing and observability. Pricing appears to be not yet defined in the public materials.
Pros
- Works with popular SDKs like OpenAI, Anthropic, and Vercel AI, so swapping models doesn't require rebuilding your application
- Private model connectivity addresses a real security concern; your self-hosted models don't need public endpoints
- Fallback and key-selection failover are configurable through the SDK by passing a models array or through gateway settings
- The team is responsive to feedback; they've already added more detailed usage breakdowns and published guides for coding agents like Cursor, OpenCode, Pi, and Zed
Cons
- Fallback behavior can silently change response quality; a customer-facing reply that falls back to a weaker model reads fine but is slightly worse, and the gateway only flags it in the dashboard rather than in the response itself
- Timeout retries may not be safe for all request types; the team acknowledged that a do-not-retry attribute on requests is something they should implement, but it's not available yet
- It's not well suited for teams that need automatic model selection by category; grouping models by responsiveness, fidelity, or task type is on the roadmap but not shipped
ngrok AI Gateway fits developers and platform teams that want to route across multiple providers without maintaining their own gateway infrastructure. The self-hosted model connectivity is the most distinctive piece, particularly for teams running models on local hardware or private clouds. If you're already using ngrok's tunnel product, the workflow will feel familiar. Teams that need advanced routing policies or request-level retry controls may want to wait for the roadmap items to land.
Open 'ngrok AI Gateway' Website
Your membership also unlocks:
