Introducing Ask TFY: A New Way to Understand and Control Your AI in Production

Every second, you have AI making decisions. Agents are talking directly to your customers, routing requests, executing tools, and spending real money. The system is running, and most of the time, you have no idea what's actually happening inside it.

That's not a knock on your team. It's just the reality of running AI at scale. The data is there, in your traces, your configs, your logs, but getting to it means knowing exactly where to look, what to filter, and how to connect the dots across half a dozen different screens. When everything is working, you don't think about it. When something breaks, suddenly you're context-switching between your monitoring dashboard, your guardrail config, your routing rules, and your documentation, trying to piece together what went wrong before someone escalates it into a bigger problem.

This is where most teams are today. And it's exhausting.

What if you could just…ask?

Ask TFY is a new way to interact with your AI gateway on TrueFoundry. It's not a search bar, and it's not a chatbot that regurgitates documentation at you. It's an agent built within TrueFoundry, with live access to everything running inside your gateway: your models, your traces, your MCP servers, your guardrail configurations, your budget rules, and your data access policies. When you ask it something, it comes back with an answer grounded in what's actually happening in your system right now. 

Debug failures in seconds, not hours

Here's a scenario that might sound all too familiar. Your application starts throwing errors. You don't know if it's a model issue, a routing issue, a budget cap that got hit, or something else entirely. So you start digging, manually filtering traces, looking for patterns, checking configs one by one.

With Ask TFY, you describe the problem instead.

"My application has been failing intermittently for the last 7 days. What went wrong?"

Ask TFY goes into your trace history, classifies the errors, identifies whether failures were client-side or caused by something like an upstream model provider going down, and surfaces the pattern. It tells you what broke and why, specifically, not generically. And then it tells you what to do about it.

"How do I fix the configuration to make this more resilient?"

It recommends the right solution for your setup. Maybe that's configuring a virtual model with a fallback. Maybe it's adjusting your routing rules. Because it knows your system, it gives you an answer that actually applies, not a generic troubleshooting guide.

And then you can tell it to act on that recommendation directly.

"Apply the recommended budget and fallback configuration to that agent."

That's it. You stayed in one place, asked three questions, and the problem is resolved. No copying YAML from one screen and pasting it into another. No tab switching.

Surface insights you would never think to look for

Debugging is one use case, but Ask TFY is just as useful when nothing is visibly wrong. The most expensive production problems are the ones that quietly drain your resources or create security exposure before anyone notices.

For example, you can ask how well your provider-level prompt caching is actually being utilized. It won't just give you a number. It will tell you which specific requests are missing the cache and exactly where you need to add the cache block to fix it. That's the kind of insight that would take a skilled engineer an hour to extract manually, and most teams never bother because it doesn't feel urgent enough until the bill arrives.

Or ask about cost. Give it a time range, tell it how you want to break down spend, by developer, by application, by model, and it will generate the analysis and propose a budget configuration you can apply directly.

Why this is different

The questions that matter most in AI operations are rarely the ones that standard dashboards are built to answer. Observability tools show you what happened. Ask TFY helps you understand why, decide what to do, and then do it, without leaving the conversation.

There's no context to rebuild between tabs, no manual cross-referencing, no translating what you found in monitoring into what needs to change in config. Ask TFY holds all of that at once and works across it on your behalf.

It also knows more than just your data. Ask it something that goes beyond your gateway, like how a new third-party library integrates with your setup, what a particular model provider's API expects, or how to structure a configuration you've never set up before, and it will pull from TrueFoundry's documentation and se arch the web to fill in the gaps.

The next time you need to understand what your AI is actually doing in production, or need to act on it fast, just Ask TFY.

Ask TFY is available now in TrueFoundry AI Gateway. Sign up now to get started.