Tired of juggling 5 different LLM API keys for my side project - found a simpler way

I'm building a small AI app that uses different models for different tasks - DeepSeek for technical reasoning, MiniMax for long context documents, Kimi for conversational responses. The API key juggling was driving me crazy. Each provider has their own dashboard, billing system, rate limits, SDK quirks. For a solo dev side project, this overhead was ridiculous. A friend pointed me to novapai.ai last month. It's a token relay platform - one API key, one OpenAI-compatible endpoint, access to multiple models including DeepSeek V4 Pro, MiniMax M3, and Kimi 2.6. I was hesitant. Smaller relay platforms can disappear overnight, API latency is always a concern. After testing for about two weeks on my side project: latency is acceptable, pricing is straightforward per-token pay-as-you-go with no surprises, the OpenAI-compatible format made integration trivial (literally just changed the base URL and model name), and uptime has been fine for non-critical use. Is it perfect? No. You're adding a middleman so if they go down all your models go down. Rate limits can kick in during peak hours. But for a solo dev who doesn't want to deal with 5 different billing dashboards? It's been worth the convenience. Curious if others here use relay platforms or prefer going direct to each provider. What's your LLM API setup look like? (Not affiliated, just sharing something that worked for my workflow.)