A new API proxy called Lexi wants to cut your AI API bills with a simple pitch: change one URL in your code, and if Lexi doesn't save you money, you pay nothing.
The setup takes about two minutes. You swap your API endpoint from your provider's URL to Lexi's. Your API keys still go directly to providers - Lexi never holds them. The service currently supports 30+ models across OpenAI, Anthropic, Google, Mistral, Groq, and xAI.
How It Works
Lexi reduces costs primarily through context optimization - compressing the tokens (the chunks of text AI models process and bill you for) sent in each request while maintaining coherence. Every API response comes back with HTTP headers showing the exact cost breakdown: what the request would have cost directly, what it actually cost through Lexi, the compression ratio, and original versus compressed token counts. Full transparency on every single call.
The pricing model is performance-based. Lexi takes 40% of whatever savings it generates on each request. If a request can't be optimized, you pay the standard provider rate with zero markup. The company's explicit guarantee: you cannot pay more than going direct.
Streaming, tool calls, and structured output all work through the proxy. New accounts get $10 in credit with no card required.
The Trade-Off to Watch
Context compression always involves a quality trade-off. Lexi claims "conversations that go further" with improved coherence over long exchanges, but any token reduction means some information is being removed or condensed before reaching the model. For short, simple prompts the savings will be minimal. For long conversations or large context windows where you're feeding in entire documents, the savings could be meaningful - but so could the quality impact.
The 40% revenue share is aggressive. If Lexi saves you $100 on your monthly API bill, it keeps $40. That's fair if the alternative is paying full price, but it means the service is most compelling for teams with substantial API spend where even a net 60% of savings adds up to real money. Solo developers spending $20/month on API calls probably won't notice the difference.