Lite core and extras¶

As of 1.0.0, pip install hyperion-sdk installs a slim lite core. Heavy backends are opt-in extras. This page explains the split and the reasoning.

What is always installed¶

The lite core pulls only light, pure-Python-ish dependencies: loguru, pydantic, httpx, python-dateutil, python-dotenv, env-proxy, cachetools, click, haversine, aws-lambda-typing.

That is enough for the domain layer, the ports, configuration, logging, date utilities, async utilities, Haversine geo math, and the in-memory / filesystem adapters.

The extras¶

Extra	Pulls	Enables
`[aws]`	`boto3`, `aioboto3`	every `adapters/*/dynamodb`, `s3`, `sqs`, `aws_sm` (DynamoDB cache/keyval, S3 storage/schema, SQS queue, AWS Secrets Manager)
`[data]`	`polars`, `pandera`, `numpy`	`hyperion.data.*` (pandera↔polars typing, asset schemas, `SpatialKMeans`)
`[catalog]`	`fastavro`	`hyperion.adapters.serialization.avro` (avro-backed `Catalog`)
`[geo]`	`googlemaps`	`hyperion.adapters.geocoder.google` (`GoogleMaps`); Haversine math stays lite
`[snappy]`	`python-snappy`	compressed filesystem cache / DynamoDB keyval (graceful no-compression fallback when absent)
`[all]`	union of the above	parity with the pre-1.0 full install

[catalog] no longer transitively requires [aws]: a Catalog backed by local filesystem storage works with [catalog] alone.

Why split it¶

Smaller, faster installs — most consumers don't need boto3 and polars and googlemaps. A source running in a constrained Lambda or a CLI tool only pays for what it uses.
Clear capability boundaries — the extras map one-to-one onto adapter families, so "what do I need to install?" has a mechanical answer.
Architecturally enforced — the lite-core promise (the core pulls no heavy third-party) is verified separately by the import-graph test and a no-extras smoke job, complementing the internal layering contract described in Ports and adapters.

Picking extras¶

Install the capabilities you use, e.g.:

pip install 'hyperion-sdk[catalog,aws]'   # Catalog on S3
pip install 'hyperion-sdk[data]'          # feature models / asset collections
pip install 'hyperion-sdk[all]'           # everything (pre-1.0 parity)

See the task-focused Install Hyperion and pick extras guide for combinations and version pinning.