Deploy commands (APIPod engine)

Run socaity scan, socaity build, and socaity start after pip install socaity. These commands delegate to APIPod, the packaging engine behind the decorator. For login, service install, and day-to-day SDK use, see the Socaity CLI reference.

Install with pip install socaity. You also need the apipod Python package in your service project for from apipod import APIPod; the socaity executable handles scan, build, and start.

Installation

pip install socaity apipod

Scan

Inspect your project and generate a deployment configuration. APIPod reads your requirements.txt, detects CUDA/PyTorch/TensorFlow dependencies, and infers the correct base Docker image.

# Scan the project and generate deployment config
socaity scan

Build

Generate a Dockerfile and build your container. APIPod selects compatible CUDA/cuDNN versions automatically. Pass a target file to override the default entry point.

# Build with auto-detected settings
socaity build

# Specify the target entry-point file
socaity build ./main.py

Start

Run your service locally. Combine with --compute serverless --provider localhost to enable the in-memory job queue, which emulates serverless GPU behaviour for testing.

# Start on the default port (8000)
socaity start

# Start on a custom port
socaity start --port 9000

# Enable local job queue (emulates serverless GPU)
socaity start --compute serverless --provider localhost

Deployment Flags

These flags control which runtime backend APIPod selects. They work with both socaity build and socaity start.

# Build for Socaity orchestration
socaity build --orchestrator socaity

# Build for RunPod serverless
socaity build --compute serverless --provider runpod

# Start locally with serverless job-queue emulation
socaity start --compute serverless --provider localhost

Flag Reference

Flag	Argument	Description
`scan`	`(none)`	Scan project, detect dependencies, and generate deployment config.
`build`	`[FILE]`	Build Docker container. Optionally specify the target entry-point file (defaults to main.py).
`start`	`(none)`	Start the service locally.
`--orchestrator`	`socaity \| local`	Orchestration platform. Use "socaity" for Socaity-managed deployment.
`--compute`	`dedicated \| serverless`	Compute mode. "serverless" enables the job queue; "dedicated" runs a standard FastAPI server.
`--provider`	`localhost \| runpod \| auto`	Infrastructure provider. "localhost" activates in-memory job queue for local testing.
`--region`	`string`	Accepted but not yet applied. Region is platform-selected today (roadmap).
`--host`	`string`	Host address to bind to when starting locally (default: 0.0.0.0).
`--port`	`int`	Port to listen on when starting locally (default: 8000).

Orchestrator × Compute × Provider Matrix

The combination of --orchestrator, --compute, and --provider determines which runtime APIPod activates.

--orchestrator	--compute	--provider	Runtime
`socaity`	`serverless`	`localhost`	FastAPI + local job queue (test mode)
`socaity`	`serverless`	`runpod`	RunPod serverless router
`socaity`	`dedicated`	`localhost`	FastAPI + local job queue (test mode)
`socaity`	`dedicated`	`auto`	FastAPI
`local`	`serverless`	`localhost`	FastAPI + local job queue
`local`	`serverless`	`runpod`	RunPod serverless router
`local`	`dedicated`	`localhost`	FastAPI

Providers