// the take
APIPark is an API gateway with a built-in LLM proxy layer — you configure your AI provider credentials once and it normalizes requests across OpenAI, Anthropic, Gemini, DeepSeek, and 100+ others behind a single endpoint. It also ships a developer portal for internal API sharing, subscription/approval workflows, and usage dashboards. The target is teams that want to centralize AI provider access rather than scatter API keys across every service.
The unified request format is genuinely useful — swapping models or rotating providers doesn't require touching application code. The provider YAML config system is well-structured and easy to extend, which matters when new models ship weekly. The subscription/approval workflow is a real feature that most homegrown proxies skip, giving platform teams actual control over who calls what. One-command Docker deployment gets you running fast without wading through Kubernetes manifests.
The Go backend is the real gateway, but the repo leads with TypeScript in the language field — that mismatch creates confusion about what you're actually deploying. The model runtime provider files are mostly frozen around late-2024 model versions (no Claude 3.5 Sonnet v2, no Gemini 2.5, outdated Bedrock listings), so you'll be patching YAML files immediately after install. The README leans heavily on screenshots and a partner list that reads like a marketing page, while actual configuration docs live on an external site — finding how to configure TLS, set up clustering, or tune the Go HTTP layer requires leaving the repo. No mention of how the 'outperforms Nginx' performance claim was benchmarked.
We read the README, the directory tree and a key source file in July 2026. Nobody pays to be listed. How we judge →