EidosStack AI Gateway

The single OpenAI-compatible endpoint and per-key spending limits solve a real operations problem. Do you expose the selected upstream provider and applied token rates in each response or usage log? That would make cost audits and fallback debugging much easier, especially when routing workloads by price.

Per-key spending limits are a useful safeguard when prototyping across model families. Are those limits enforced before a request begins, and can an in-flight streaming response exceed the remaining budget? Documenting that behavior would help developers set safe limits for browser demos.

With token prices refreshing every 30 seconds, is a request billed at the price when it starts, even if a long streaming response crosses a price update? It would help to see the applied rate and separate input, output, and cache charges in a per-request usage record.

The live pricing and per-key spending limits are practical for teams comparing providers. Is there an SDK or request-level fallback example for switching models while keeping the same OpenAI client code?

I built this because comparing token prices across providers was a pain. Every provider has a different pricing page, and by the time you finish doing the math you forget which model you were checking. So I put live input, output, and cache pricing side by side on one page, refreshed every 30 seconds, and everything is callable through a single OpenAI-compatible endpoint. Prices run about 50% below official rates since we buy capacity differently. Would love feedback on the pricing table, and happy to answer questions about how the routing works.