Skip to content

Superb AI Python SDK

The official Python client for the Superb AI platform — the full MLOps loop: datasets → labeling → versioning → training → deployment, all from typed Python.

View on PyPI{ .md-button }

pip install superb-ai

from superb_ai import Client

client = Client(tenant="acme", api_key="sbd_pk_...")   # or SUPERB_AI_TENANT + SUPERB_AI_API_KEY
projects = client.projects.list()                       # you're in

What's here

  • Quickstart — auth, then a full COCO dataset imported end-to-end.
  • Recipes — twelve task-shaped walkthroughs (jobs, workflow, versions, exports, search, training, deploy/predict, auto-label, …).
  • Limits & quotas — rate limits, concurrency caps, resource quotas, size caps, and the gotchas worth knowing before you hit them.
  • API reference — every namespace, method, argument, return type, and the typed error each can raise — generated from the code.

Design in one screen

  • Sync + async twins. Every client.x.method(...) has an awaitable counterpart on the async client; identical signatures.
  • Typed errors. Exceptions derive from SuperbAIError and carry a stable .code (an ErrorCode) plus a human .hint — branch on .code, never on message text.
  • Page[T] pagination. List methods return a page you iterate to auto-fetch every page; include_total=True adds .total.
  • JobHandle for async work. Kickoff methods return a handle — call .wait(); it raises JobFailedError / JobTimeoutError and exposes .result_summary.
  • Runtime deps are just httpx + pydantic. Python ≥ 3.12.

Working with an AI agent?

The package ships an AGENTS.md skill doc and an api-map.md method index designed to be read by coding agents. Point your agent at them.