Concepts

Architecture

The three layers of the stack: the Socaity SDK that calls a model, APIPod that wraps your own code as an endpoint, and the orchestrator that places the work on a GPU.

What is MaaS

Model-as-a-Service: one API key, one Python import, and a hosted model behind it. Covers what MaaS includes, what it does not, and when to use Socaity over a raw provider.

Job System

A Job is one asynchronous invocation of a model. This page lists the nine statuses (PENDING, QUEUED, PROCESSING, STREAMING, FINISHED, FAILED, TIMEOUT, CANCELLED, UNKNOWN) and the polling defaults that move a Job between them.

Serverless vs Dedicated

Serverless GPU scales to zero and bills only for execution seconds. Dedicated GPU runs always-on at a fixed monthly rate with no cold start. This page is the decision tree between the two.

Cold Starts

A cold start is the time between a request landing on an idle endpoint and the first worker becoming ready. Covers what drives it, when it matters, and the levers Socaity pulls to keep it under four seconds.

Pricing Model

Active-only billing means you pay for GPU-seconds in PROCESSING state. You do not pay for idle, queued, failed, or cancelled jobs. This page walks through the line items and the pass-through provider rates.

Cloud Providers

A provider is the GPU vendor that runs the workload. APIPod runs two today, localhost and RunPod, with Scaleway and Azure planned.

EU Hosting

Socaity runs workloads on EU-native infrastructure (RunPod EU today; Azure North/West Europe planned) and lets you pin a region per endpoint. This page covers data residency, GDPR alignment, and the regional picker.

Aliases and resolution

How the SDK turns a Python import like flux_schnell into a live endpoint URL. Covers official vs Replicate-backed aliases, the resolution flow on first call, and what happens when an alias is unknown or has no active deployment.