Back in September, I promised some "og ngrok" stuff that we'd been cooking, and some of it's real now, like observability for all your TCP and TLS connections. Meanwhile, I'm genuinely having trouble keeping up with everything our AI Gateway team is shipping… I keep telling myself it's a good problem to have.
Observe your TCP connections… finally. If you've been using ngrok to connect services that carry Postgres, MQTT, or end-to-end encrypted TLS, Traffic Inspector never had anything to show you. Now every TCP and TLS connection shows up with its IP and location, duration, bytes in/out, and why it closed. We don’t read the bytes themselves (for good reason), so it’s metadata only.
Know your limits before they bite. In the olden days, your Limits page showed a list of numbers. That was all. Now it shows you exactly how close you are to each limit, whether each one is Good, Approaching Limit, or Limit Reached, and when a limit last turned away requests. Many of you asked for this, and even more limits are landing over the next few weeks.
Plop a firewall between your apps and your models. ngrok.ai's new Firewall checks what your apps send to models and what comes back to block anything risky or out of place. Attach a policy to your keys so every app and harness gets the same rules at once. And remember Jev? Didn't that feel like a year ago? Well, Jev powers the rules you write in plain English, like "deletes files outside the project directory."
Name and define models however you want with virtual models. Give an ordered list of ngrok.ai models one easy name, like rubber-duck, and your apps try each model in order. If Claude goes down, fall back to Gemini. If Gemini goes down, flop over to a third. And you can change it once in the dashboard without redeploying or making any changes in your app's code.
Okay, enough product news. If you read one deep-dive thing from us this month, make it this one. I wrote it, so yes, I'm biased.
What the heck is the Public Suffix List? It's how your browser knows where one website ends and the next begins. Mozilla built it to keep cookies safe, and now password managers, certificate authorities, and even ngrok own guts rely on it. Multi-tenant platforms (like ngrok, again) ballooned its size and importance, and after nearly 20 years, nobody has built a better alternative. I asked two volunteer maintainers where the list goes next, and they gave me quite different answers about what happens to this consequential 0.3 MB of plaintext.
We also:
- Launched
@ngrok in Slack, which I teased last month. Connect your account and ask about your endpoints, traffic, usage, and limits right from a channel.
- Shipped official ngrok packages for Next.js, FastAPI, Axum, and Gin, which open a public endpoint when your app starts and close it when your app stops.
- Taught AI Gateway to generate and edit images (Gemini, OpenAI, xAI), handle voice in and out (Gemini, OpenAI, Qwen Omni), and create embeddings.
- Brought GPT-6.1 Sol, Meta's Muse models, Claude Haiku 5.5, Gemini 3.1 Pro, Step 5, and the Gemini Interactions API to AI Gateway.
- Gave slow-thinking models up to 15 minutes per request, up from one minute, and made the gateway tell your app when an answer gets cut off.
- Shipped automatic top-ups for AI Gateway credits, and launched status.ngrok.ai so you hear about problems from us first.
There's the October haul—and, yes, I know it's November now. I'll be back in just a few weeks, before the Thanksgiving holiday in the U.S., with how your coding agents are about to get a hell of a lot better at driving ngrok.
Until then, hit reply and tell me: What kind of questions do you wish you could just ask about your traffic? Or your ngrok infrastructure at large?
|