Foglamp came out of a problem Gustavo Fior kept hitting: he was shipping AI agents he couldn't see, costs creeping up with no idea which agent was to blame, one quietly looping for an hour and burning tokens. So he built open-source observability for agents on the Vercel AI SDK. Wrap your model in one line and every call reports cost, latency, tokens, full traces, and alerts.
🔥 Our Take: You know how everyone spent the last year building AI agents? The bills are starting to show up, and it turns out most people have no real idea where the money is going. An agent gets stuck in a loop overnight, burns through tokens until morning, and you you're landed with an invoice. That's literally why Gustavo built this, it happened to him. There are already tools in this space, Langfuse and Helicone, so he isn't first. But Foglamp is built for the AI SDK everyone is already using, and you can have it running in about two minutes. With this kind of thing the whole game is whether the setup is annoying enough that you give up before you see the value. This one isn't.