cascades-sdk 0.3.1

Python SDK for Cascades: capture-mode @task/@flow DAG compiler and thin HTTP client for the control plane

Cascades SDK (Python)

Official Python client for Cascades — a governed execution system. This project is licensed under the Business Source License 1.1 (BUSL 1.1); see License. The package provides capture-mode authoring (@task / @flow) to build a serializable DAG, plus an HTTP client aligned with the mirrored OpenAPI contract.

PyPI: cascades-sdk · Repository: no1rstack/cascades-sdk · HTTP contract (mirrored): contracts/api.yaml (source of truth: no1rstack/cascades)

What this library is (and is not)

In scope	Out of scope (handled by the Cascades platform / your deployment)
@task / @flow decorators and capture-mode execution to record a DAG	Worker execution, retries, queues, scheduling
Deterministic DAG JSON (build_dag_from_flow, canonicalization helpers)	Routes omitted from contracts/api.yaml (see platform /openapi.json for the full app)
CascadesClient — every path in contracts/api.yaml (workflows, runs SSE, integrations, system version)	Auth0 login UI, Stripe webhooks, internal admin routes
Auth helpers (SessionCookieAuth, CookieHeaderAuth, HeaderAuth, CompositeAuth) matching contract security	Issuing session cookies (browser / Auth0 flow)
wait_for_completion / wait_for_completion_async — block on SSE (GET /api/runs/{id}/stream) until a terminal run status	In-process orchestration engine inside this wheel

The SDK does not ship Airflow/Argo/BPMN adapters, W3C PROV builders, CloudEvents emitters, or OpenTelemetry instrumentation — those belong in documentation or separate packages if you add them later.

Install

pip install cascades-sdk

Import namespace: cascades_sdk (not cascade_sdk). Legacy wheel name noirstack-cascade-sdk is obsolete; use cascades-sdk only.

Contract version parity

cascades_sdk.API_CONTRACT_VERSION must match info.version in contracts/api.yaml. When the platform bumps that field, mirror the YAML and update src/cascades_sdk/_meta.py before releasing (tests/test_contract_parity.py enforces this).

Quick start

from cascades_sdk import CascadesClient, SessionCookieAuth, wait_for_completion
from cascades_sdk.compiler import build_dag_from_flow

# Session value from your Auth0-backed deployment (__session by default).
client = CascadesClient(
    "https://your-cascades-host",
    SessionCookieAuth("paste-session-cookie-value-here"),
)

version = client.get_public_api_version()  # GET /api/system/version (no auth required)
# Response dicts use the contract's camelCase keys (e.g. runId), not snake_case — same JSON as OpenAPI.
accepted = client.submit_workflow_run({"workflowId": "your-catalog-id"})
run_id = accepted["runId"]

# Blocks on SSE until SUCCESS / FAILED / CANCELLED (see contracts/api.yaml Runs tag).
terminal_event = wait_for_completion(client, run_id, timeout=3600.0)
print(terminal_event)

Async: from cascades_sdk import wait_for_completion_async and await wait_for_completion_async(client, run_id).

Optional header auth (only if your tenant documents it): CompositeAuth(SessionCookieAuth("..."), HeaderAuth({"X-Custom-Token": "..."})).

Optional product markers: pass vendor_headers=True to CascadesClient to add stable X-SDK-* / X-Product / X-Vendor headers (see vendor_telemetry_headers() in cascades_sdk._meta). The default User-Agent already names the SDK, Cascades, repo URL, and Noir Stack LLC.

Aliases: CascadeClient and CascadeSDKError are kept for backward compatibility; prefer CascadesClient / CascadesSDKError.

Capture mode and data at task boundaries

Under @flow, task calls are intercepted so the compiler can record nodes and edges without running your real side effects on the machine where you compile. Values that end up in the DAG should be JSON-friendly (e.g. str, int, float, bool, dict, list, None). Avoid pushing live handles (DB connections, open files, arbitrary class instances) across task boundaries if they cannot serialize the way your control plane expects.

build_dag_from_flow(flow, flow_inputs=None) — pass flow_inputs when the flow function has parameters you need to bind for capture.

HTTP API surface

The client mirrors contracts/api.yaml only (16 paths), including:

• get_public_api_version() → GET /api/system/version
• submit_workflow_run(body) → POST /api/workflows/run
• clone_catalog_workflow(body) → POST /api/workflows/clone
• iter_run_stream_events(run_id, since=..., task_since=...) → GET /api/runs/{id}/stream (SSE RunStreamEvent payloads)
• save_*_integration / test_*_integration → POST /api/integrations/...

JSON keys match the deployment and contracts/api.yaml (camelCase). That is deliberate for a thin client: payloads align with the HTTP contract rather than idiomatic Python dict style. Typed shapes live under cascades_sdk.types for common response bodies.

HTTP contract mirror (maintainers)

Canonical upstream: no1rstack/cascades → contracts/api.yaml. Do not hand-edit the copy here; sync and verify. See contracts/README.md, RELEASE.md, and:

./scripts/sync_contract.sh ../cascades/contracts/api.yaml
# or: make sync-contract

Publish to PyPI (maintainers)

Preferred: GitHub Actions → Publish to PyPI (workflow_dispatch) using a trusted publisher — see RELEASE.md for the one-time PyPI UI steps (no PYPI_* secret stored in GitHub).

Local / token file: build, check, then upload with Twine:

python -m build
python -m twine check dist/*
python -m twine upload dist/*

On Windows, load variables from a .env-style file (default: ..\cascades\.env.local; override with -EnvFile or CASCADES_SDK_ENV_FILE):

powershell -ExecutionPolicy Bypass -File .\scripts\publish_pypi.ps1
# or:  ... -EnvFile "C:\path\to\.env.local"

Use PYPI_USERNAME + PYPI_API_KEY, or TWINE_USERNAME + TWINE_PASSWORD, or PYPI_TOKEN / PYPI_API_TOKEN (script maps token to __token__ + password). Optional PYPI_URL: use an upload endpoint (e.g. https://upload.pypi.org/legacy/ or TestPyPI’s upload URL). Do not use a project page URL (https://pypi.org/project/...).

Publish to npm (maintainers)

Optional scoped npm package @noirstack/cascades-sdk (mirrored contracts/api.yaml, LICENSE, ESM index.js / index.d.ts) — the tarball is limited by the files array, and prepack / prepublishOnly run npm run build before pack/publish. See packages/cascades-sdk/ and RELEASE.md. PyPI cascades-sdk remains the Python client.

Try a tarball from the repo root (requires npm install once so the workspace resolves):

npm run pack:npm -- --dry-run
# equivalent: npm pack -w @noirstack/cascades-sdk --dry-run

Or cd packages/cascades-sdk and run npm pack / npm pack --dry-run as usual.

CI: Actions → Publish npm @noirstack/cascades-sdk (requires NPM_TOKEN repository secret).

Local: same default env file as PyPI (..\cascades\.env.local, or CASCADES_SDK_ENV_FILE / -EnvFile). Set NPM_TOKEN or NODE_AUTH_TOKEN.

powershell -ExecutionPolicy Bypass -File .\scripts\publish_npm.ps1
# Dry run:  ...\publish_npm.ps1 -DryRun

Contributing

Issues and PRs: github.com/no1rstack/cascades-sdk. For compiler edge cases, include a minimal @flow / @task snippet and the DAG you expected vs what build_dag_from_flow produced.

License

This project is licensed under the Business Source License 1.1 (BUSL 1.1). See LICENSE for full terms.

Non-production, evaluation, and internal research use are permitted under the Additional Use Grant in LICENSE. Commercial use, SaaS deployment, or offering the Licensed Work as a service requires a commercial agreement with Noir Stack LLC — https://noirstack.com. On the Change Date (see LICENSE; 2030-05-01, subject to the four-year rule in the license text), the Licensed Work converts to GPL-3.0-or-later.

The Cascades SDK is part of the Cascades system and follows the same licensing model as the platform unless otherwise explicitly stated.