xk1ko/aigloo ? reverse-engineered prompt

Build me a self hosted AI gateway that I can run locally or in Docker, with one endpoint and one API key for my tools. I want to point apps like Cursor, Claude Code, Codex, and other clients that support a custom base URL at it, and have it translate between provider formats, route requests to different models, and automatically fall back if a provider is slow or fails.

It should include a clean web dashboard where I can add provider keys, choose models, create access keys for other people or devices, and set limits like allowed models, request rate, and rolling spend caps. I also want usage tracking, budgets, alerts when limits are hit, and a way to see spend by access key. If possible, add a few token saving options that help trim or compress context before sending requests. Make the setup easy, with a quick first run, sensible defaults, and a config file I can edit later.