PyFlumeNG

PyFlumeNG is a fork and superset of PyFlume. It uses the dedicated pyflumeng import package and retains the legacy helpers while adding the customer-portal OAuth flow, portal write operations, typed resource models, pagination, response/error handling, rate-limit state, and the endpoint surface used by Flume's customer portal.

Install PyFlumeNG from PyPI:

pip install PyFlumeNG

Authentication

PyFlumeNG supports two ways to obtain a Flume bearer token. They are not tied to separate sets of read helpers: ordinary reads such as devices, notifications, leaks, usage alerts, and queries can be used with either auth object. The main difference is how the token is obtained and what scopes Flume grants it.

Auth class	Credentials	Best fit	Portal-only writes
PersonalAuth / FlumeAuth	Email, password, API client ID, API client secret	Flume's documented Personal API flow	No
PortalAuth / FlumePortalAuth	Email and password	Customer-portal flow and code that needs the portal's broader token	Yes

PersonalAuth and FlumeAuth are the same class. Use them when you want the documented Personal API password-grant flow with credentials generated from Flume's API Access settings:

import pyflumeng

auth = pyflumeng.PersonalAuth(
    username="your_email",
    password="your_password",
    client_id="your_client_id",
    client_secret="your_client_secret",
)
client = pyflumeng.FlumeClient(auth)

PortalAuth and FlumePortalAuth are the same class. This flow uses the same customer-portal OAuth path as Flume's web app and needs only the account credentials:

import pyflumeng

auth = pyflumeng.PortalAuth(
    username="your_email",
    password="your_password",
)
client = pyflumeng.FlumeClient(auth)

You can use that PortalAuth object for normal reads too. For example, client.list_notifications(), FlumeNotificationList(auth), client.list_devices(), and FlumeData(...) all accept it. Portal auth becomes required when an operation needs a scope that Flume does not grant to a normal Personal API token, including the portal usage-alert rule write routes.

See Authentication for the full distinction, including which auth mode to choose and how the legacy helpers behave.

Typed API

Named resource methods return typed models. For example:

user = client.get_user()  # Optional[User]
devices = client.list_devices()  # list[Device]
rules = client.list_usage_alert_rules("device")  # list[UsageAlertRule]
accuracy = client.get_meter_accuracy("device")  # list[AccuracyResult]

Models expose known API fields to type checkers and IDEs while remaining forward-compatible with Flume additions. Unknown response fields remain available as attributes or mapping keys and are preserved by to_dict().

Higher-level typed helpers are available for common dashboard data as well: get_usage_breakdown() combines classified spans with a whole-window SUM query, rolls raw irrigation into Outdoor, derives Indoor as the unclassified residual, and returns category usage/percentages. Budget exposes used, target, remaining, percentage_used, and is_over_budget progress properties. Portal semantic helpers additionally cover notification preference bitmasks without dropping unknown bits, structured usage-alert history, AVG notification detail, validated usage-rule/Smart-Leak and Do Not Alert payloads, and percentage-based budget configuration.

Custom usage-alert validation includes the live API rule that repeat notifications must be at least twice the alert duration. If callers omit notify_every, PyFlumeNG uses the safe minimum 2 * duration automatically.

pyflumeng ships a py.typed marker so installed type checkers can consume the package annotations. The complete method signatures and model fields are in the generated API reference.

Endpoint coverage

FlumeClient includes the documented Personal API plus portal-discovered routes for users, devices, current flow and queries, locations and location profiles, notifications, usage alerts and rules, do-not-alert schedules, budgets, subscriptions, emergency contacts, shared location access, integrations and shutoff valves, spans and span types, feedback, meter accuracy, support conversations, purchase options, insurers, professional services, API clients, contacts, and subscription/Stripe portal operations.

List helpers follow Flume pagination and return all pages. response() exposes the complete typed response envelope when pagination or metadata matters, and raw() remains an escape hatch for routes added by Flume after this release. See the FlumeClient guide for the main interface.

Customer-portal responses have been observed advertising a 72,000-request quota through X-RateLimit-Limit. The accompanying reset epoch varies between live requests and should not be treated as a fixed midnight reset. FlumeClient.rate_limit tracks the live server-reported limit, remaining requests, reset time, and Retry-After value so frequent pollers can budget requests from the current quota state. See Authentication for details.

Legacy helpers

The original helper classes remain available and typed:

• Data retrieval
• Devices
• Leak alerts
• Notifications
• Usage alerts

Examples

The examples include an interactive portal utility that lists usage-alert rules and explicitly enables or disables the selected rule.

Development

The API reference is generated from source signatures and model annotations:

python scripts/generate_api_reference.py
python scripts/generate_api_reference.py --check

tox runs the tests, mypy, generated-reference check, and bytecode compilation.