One OpenAI-compatible key for 100+ AI models - how and why we built Mervex

We are a tiny bootstrapped team (zero funding, working from home). We kept paying for - and juggling - separate AI API keys for every model we needed: one for Llama, one for Qwen, one for embeddings. So we built our own gateway: Mervex.\n\n## What it does\n\n- One OpenAI-compatible key -> 100+ models (Gemma, Llama, Qwen, GLM, Kimi, DeepSeek and more)\n- Drop-in for your tools: Open WebUI, LibreChat, Chatbox, Cursor - you literally change the base URL and swap the key\n- Auto-Decision API - a weird little extra we love: millisecond probabilistic decisions (yes/no, option picking, scoring, even image checks) with 10k free lifetime calls\n- Live public monitoring per model - success rate, latency, 14-day history. If a source dies, requests fail over automatically\n\n## Free tier (no credit card)\n\n- 100 requests daily\n- Bonuses: connect Discord / Google / GitHub for extra daily quota\n- One-time +400 welcome bonus\n\n## Why build this instead of using a big provider?\n\nHonest answer: learning + independence. Big gateways are great until a model disappears, pricing changes overnight, or your region gets restricted. A gateway YOU control can fail over between sources and keep serving. We wrote about that philosophy more on our blog.\n\n## Links\n\n- Site: https://mervex.cc.cd\n- Docs (Postman): https://documenter.getpostman.com/view/59154340/2sBYHUSzJ2\n- OpenAPI: https://api.mervex.cc.cd/api/openapi.json\n- GitHub: https://github.com/Msrfteam/Mervex\n\nIf you try it - roast it. What model should we add next? What did we get wrong?