reliafy-client — Reliafy from Python and the command line

Fit a model anywhere with SurPyval, then make it a shareable, trackable, citable artifact in Reliafy. Call any Reliafy tool from a script or a shell, send files and photos, and push data from your CMMS.

pip install reliafy-client

The package installs as reliafy-client; you import it as reliafy. It also installs the reliafy command, and the same command as reliafy-client: if some other program called reliafy is already on your path, use reliafy-client.

Pure standard library: no dependencies. You bring SurPyval to do your own fits.

Which server

Version 0.3 calls Reliafy's tools API (POST /api/v1/{tool}/{action}). It's on reliafy.com from the release that adds it; a self-hosted server needs a build that includes it (GET /api/v1/tools answers). Against an older server, the new calls stop with a clear message (reliafy.ServerTooOld), and everything from 0.2 (reliafy.push, reliafy.models.*, reliafy.data.*, reliafy.strategy.*, reliafy.fleet.*) keeps working.

Quick start

import surpyval as sp
import reliafy

# Fit however you like: censored data, offsets, whatever.
model = sp.Weibull.fit(x=failures, c=censoring_flags)

reliafy.configure(token="rlf_...")            # from Reliafy → API access (Pro)
url = reliafy.push(model, name="Pump bearings — 2026 refit", unit="hours")
print(url)                                    # open it in Reliafy

push() uploads the model and its fitted data by default, so Reliafy shows the full probability plot with confidence bounds, and the model stays editable and refittable in the app, and can be cited as RCM evidence, used in RBD blocks, or tracked against a fleet. Offset, limited-failure-population and zero-inflated fits are detected from the model and reproduced.

# Parameters only (no data upload): functions and life metrics, no plot.
reliafy.push(model, name="From a report", data=False)

# No SurPyval object, just numbers:
reliafy.push_params("weibull", [1200.0, 2.3], name="Handbook value", unit="hours")

Every tool: Client.call

Everything Reliafy's AI tools can do, with the same arguments and results:

from reliafy import Client

rf = Client()                                   # token from RELIAFY_TOKEN or `reliafy login`
rf.call("list", "models")
rf.call("life_data_analysis", "fit", data=[120, 340, 510, 700], censored=[0, 0, 1, 0])
rf.call("strategy", "optimal_replacement", distribution_id="weibull",
        params=[{"name": "alpha", "value": 1200}, {"name": "beta", "value": 2.5}],
        planned_cost=200, unplanned_cost=1500)
rf.call("guide", topic="life_data")             # a tool without actions
rf.call("rbd_analysis", "analyze", rbd_id="…", wait=True)   # waits for a long job

rf.tools()                                      # what you may call
rf.spec()                                       # the OpenAPI spec

Shortcuts for the common things: rf.list(kind), rf.get(kind, id), rf.fit(data, distribution=…, censored=…, save_as=…), rf.reliability(model_id, times), rf.guide(topic).

reliafy.call(tool, action, **args) does the same with the configured token. A refused call raises reliafy.ReliafyError with status (the HTTP status) and detail (what Reliafy said, in plain words).

Files, photos and data

rf.upload_dataset("bearings.xlsx", name="Bearings")      # CSV or .xlsx
rf.import_rbd("plant.xml")                               # BlockSim, Open-PSA, Galileo, RePyability
rf.import_excel("fmea.xlsx", target="rcm")
rf.add_photo("nameplate.jpg", target_id="…")             # Reliafy Maintenance

reliafy.push_usage(fleet_id, [{"name": "Truck 01", "current_use": 5230}])
reliafy.push_readings([{"tag": "PT-101", "value": 4.2}])
reliafy.push_lives(dataset_id, file="new_failures.csv")

Files are streamed from disk, not read into memory.

The command line

reliafy login                                    # saves your token (only you can read the file)
reliafy tools                                    # every tool and action you may call
reliafy <tool> <action> --arg value …            # any of them; --help on each
reliafy models list
reliafy fit failures.csv --dist weibull --censor-column status --c-invert
reliafy push model.json                          # {distribution, params | data, name?, unit?}
reliafy upload bearings.xlsx --name Bearings
reliafy import rbd plant.xml
reliafy photo add IMG_*.jpg --to POSITION_ID --caption Nameplate
reliafy ingest usage meters.csv --fleet FLEET_ID
reliafy ingest readings readings.csv

Values are read as their argument's type: text, numbers, true/false, JSON for lists and objects (--times 100,500,1000 works too), and @file reads a file. --args '{…}' gives all the arguments as one JSON object. Add --json for JSON output. reliafy photo add reports each file and exits non-zero if any failed.

reliafy fit never guesses what a flag column means: Reliafy reads censoring as 0 = failed, 1 = still running; if yours is 1 = failed, add --c-invert.

Configuration

• RELIAFY_TOKEN and RELIAFY_BASE_URL, or reliafy login, which saves them in ~/.config/reliafy/config.json (%APPDATA%\reliafy on Windows; RELIAFY_CONFIG_DIR to choose).
• Self-hosted: reliafy.configure(token=…, base_url="https://reliafy.mycompany.com"), Client(base_url=…) or reliafy login --base-url …. A local server with sign-in off needs no token.

Notes

• The token is a personal API token, created on the API access page in Reliafy. On Reliafy Cloud this is a Pro feature; self-hosted instances have it unconditionally. A token's scopes (read, write, ingest) decide what it may do.
• Supported distributions for push: Weibull, Exponential, Normal, LogNormal, Gamma, LogLogistic, Exponentiated Weibull, Gumbel, Logistic.
• The API in full: docs/api.md.
• Changes: CHANGELOG.md.