Python SDK

strangeworks-aura-sdk embeds Aura in a Python backend — a service, an API, a worker, a scheduled job — so your users get an answer from the product they already use.

It wraps the public /api/v1 surface with company API-key auth, and is async: every call is a coroutine, so it drops into an existing async service without a thread pool in the middle.

To administer Aura or try a deployment out by hand, use the CLI. There is no Python CLI; the aura command ships from TypeScript only.

Install

Python 3.11 or newer.

pip install strangeworks-aura-sdk

The import package is aura_sdk. It is fully typed and depends on httpx, pydantic (v2) and websockets.

Server-side only

An API key carries your whole company’s access. Read it from the environment or your secret store — never commit it, and never ship it to a client.

Hello, Aura

import asyncio, os
from aura_sdk import AuraClient

async def main() -> None:
    async with AuraClient(api_key=os.environ["AURA_API_KEY"]) as aura:
        project = await aura.create_project(name="vrp demo", from_deployment="acme/vrp")
        reply = await aura.chat(project.id, "Solve this and summarize the result.")
        print(reply.content)

asyncio.run(main())

acme/vrp is a deployment handle, <company>/<name> — the handle field on every deployment the SDK returns. Deploy a model is how one comes to exist.

The shape of the surface

Everything hangs off one AuraClient, whose methods group by the noun they act on:

Area	Covers
Deployments	Create, list, check and delete; deploy_from_folder uploads a folder into one
Agent instructions and skills	The operating rules a deployment ships and a project layers on top
Files	List, read, upload, delete and zip, against a deployment, workspace or snapshot
Projects and conversations	Create, list, fetch and delete
Chat	chat sends a message and returns the reply; send_message and get_conversation_state drive the wait yourself; cancel stops a turn
Run dashboard	A browser-ready URL for a project’s dashboard that carries a view ticket, not your key
Executions, snapshots and tags	What each solve produced, the workspace it froze, and movable names for the good ones
Workspace	What the live workspace sits on, and checking a snapshot or deployment back out
Usage	Credits spent — company-wide, per user, per API key

The API reference lists every method, grouped the same way. What a method returns is a pydantic model from aura_sdk.models, and the few operations the client does not wrap are listed with the REST reference.

Working with a coding agent?

This site is published as plain text: /llms.txt is the index, /llms-small.txt the guides, /llms-full.txt everything including this reference. Point Claude or Cursor at one of them.

Where to go next

ConfigurationKeyword-only arguments, the two timeouts, closing the client, and the httpx transport seam.

ConventionsAsync-only, paging, pydantic models, how errors are typed, and what the client does not wrap.

API referenceEvery export, generated from the package's own docstrings.

Chat with a projectHow the wait works, watching a turn live, and how to stop one.

Instructions and skillsGive the agent your vocabulary, rules, and procedures.