Architecture

How the Socaity SDK, APIPod, and Cloud infrastructure fit together.

The Three Layers

Socaity has three distinct concepts. Understanding the boundary between them will prevent most configuration errors.

SDK — Use

Call any hosted AI model as a Python or JS function. No HTTP, no parsing.

pip install socaity

APIPod — Deploy

Package your own AI model as a serverless API with one decorator. apipod --build → deploy.

pip install apipod

Cloud — Run

Serverless GPU infrastructure (RunPod, Socaity Cloud). Scales to zero when idle.

socaity.ai / runpod.io

Your Code

SDK call

APIPod Service

Docker

Serverless GPU

RunPod / Cloud

Result

image / audio / text

Component	Package	Role
Socaity SDK	`socaity`	Python + JS client to call hosted models.
APIPod	`apipod`	Framework for packaging your model as a cloud API.
media-toolkit	`media-toolkit`	Unified file handler — images, audio, video.
Socaity Cloud	`socaity.ai`	Managed serverless GPU runtime and model registry.
RunPod	`runpod.io`	Third-party GPU provider supported by APIPod.

Aspect	Dedicated	Serverless
Billing	Hourly, always running	Scales to zero when idle; pay-per-call when running
Scaling	Manual	Instant, any replica count
Cold Start	None	5–20s first request
DevOps	Required	None

Cold Start: Serverless containers spin down when idle. The first request after idle typically takes 5–20 seconds. Subsequent requests are fast.