Serve

Run Pie as an inference server with the APIs and models you already use.

You can run Pie as an inference server without writing any code. Start pie serve, point your existing client at it, and choose a model and hardware backend.

OpenAI SDKcompat-openaiAnthropic SDKcompat-anthropicGemini SDKcompat-geminipie serve · :8080engineone model
Each API is an inferlet that ships with Pie. They all share one port and one engine.

Serve an API

Drop-in OpenAI, Anthropic, and Gemini endpoints.

Models

Import checkpoints from Hugging Face and choose a quantization.

Hardware

Run on CUDA, ROCm, Metal, Vulkan, WebGPU, and TPUs.

Configure & tune

Edit one TOML file and tune kernels for your device.

Embed in your app

Embed Pie in Python, Node.js, browsers, iOS, Android, or C.

Scale out

Run across multiple GPUs and machines with cache-aware routing.

Go deeperEvery endpoint is an inferlet →

The OpenAI, Anthropic, and Gemini APIs are inferlets that ship with Pie. When a standard API is not enough, write your own.