EidosStack AI Gateway

Short version: one OpenAI-compatible API in front of 48+ models, with live token pricing visible before you spend anything.

The annoying part of working with LLMs was never the API calls, it was the pricing pages. Every provider formats prices differently, some bill per 1K tokens, others per 1M, cache pricing is all over the place, and by the time you've computed what a request costs, you've already made it.

So I built EidosStack AI Gateway:

• One API, 48+ models from OpenAI, Anthropic, Google, xAI, DeepSeek, Meta, Mistral, and more. If you've used the OpenAI SDK, you already know how to use it, change the base URL and key, done.

• Live pricing table: input, output, and cache prices for every model, shown before you send the request. Updated continuously instead of a static docs page.

• Up to 50% below official API rates on usage, the gateway buys capacity smartly and passes part of that on.

• One invoice instead of six, with per-model usage breakdowns.

A few things I learned building it:

1. Normalizing pricing across providers is 80% data plumbing, 20% engineering. Every provider has quirks (some include cache in input price, some don't, some bill images per pixel).

2. Retries and provider failover matter more than raw latency. The gateway retries a failed request on a healthy route and that alone saves hours of debugging.

3. Developers don't read docs pages to find pricing, they want it next to the model picker. So that's where it lives.

What's next: more models, better usage analytics, and team spend controls.

We're launching on Product Hunt on Sep 8, if the pricing-table idea resonates, I'd love your feedback there too.

Product page: indiehackers.com/product/eidosstack-ai-gateway

Site: eidosstack.com/en/ai-gateway

Ask me anything about gateway architecture or the pricing normalization mess.