AI gateway & observability
One endpoint in front of every provider — with budgets, keys and traces you own.
An AI gateway is a proxy that presents one OpenAI-compatible endpoint and routes to whichever provider you configure, adding virtual keys, per-key budgets, fallbacks and a spend log. Paired with a tracing tool such as Langfuse it also answers what a request actually did, and what it cost.
Every company that uses more than one model provider arrives at the same three problems: keys are pasted into too many places, nobody can attribute spend to a feature, and when a provider degrades, everything downstream degrades with it. A gateway solves all three without touching application code — the base URL stays the same while the routing behind it changes.
The observability half matters just as much once agents are involved. A single user request can become forty model calls, and without traces the only debugging tool is a guess. Langfuse records the tree, the latencies and the cost per step; Opik adds scoring so a prompt change can be measured rather than argued about.
The applications in this stack
Straight answers
Does this replace my provider accounts?
Can I mix local and commercial models?
Is trace data private?
Put your AI stack on your own box
Pick an app, pick a size, and have it running today. Month to month, cancel whenever.