Use Reliafy from scripts: the API and the command line

Everything you can ask Reliafy's Claude connector to do, you can also do from a script. The same tools (fit a model, analyse a diagram, plan a test, work out a replacement interval) answer over plain HTTPS, from the reliafy command line, and from Python. Use them when a job should run without anyone clicking: a fit that refreshes every week, meter readings pushed every night, results pulled into a notebook or a report.

This guide walks through the common jobs, start to finish. The API reference has every detail; each step below links to the page that covers it.

1. Create a token

A script signs in with a personal API token. Open Settings → API access, give the token a name you'll recognise later ("Nightly meter export"), and choose what it can do:

• Read: reads and calculations, the tools that change nothing.
• Write: anything that saves, changes, shares or deletes.
• Push data: the ingest routes only. A token that only pushes data can never read your analyses, which makes it the safe choice for a job on a shared server.

Press Create token and copy it: it starts rlf_ and is shown only once. Revoke it on the same page at any time. Scopes, limits and error codes are on Tokens and errors.

2. Install the command line and sign in

The command line and the Python client come in one package, with no other dependencies:

pip install reliafy-client
reliafy login        # paste the token when asked
reliafy whoami       # your account and plan

reliafy login saves the token in a file only you can read, so later commands and scripts need nothing more. On a server, set RELIAFY_TOKEN instead. (If another program called reliafy is already on your path, the same command is installed as reliafy-client.)

3. Fit a model from a CSV

Take a CSV of failure times with a censoring column, hours,censored, where 0 means the unit failed and 1 means it was still running:

reliafy fit bearings.csv --dist weibull --unit hours --save "Bearing life"

Fitting 10 rows: times from 'hours', censoring from 'censored' (0 = failed, 1 = still running).
saved: true
model_id: …
url: https://reliafy.com/modelling/m/…
…
alpha  996.3
beta   3.82

The model is saved in your workspace, with its data, exactly as if you'd fitted it in the app: open the url to see the probability plot. Leave out --save to fit without saving, and --dist to let Reliafy pick the best distribution.

Two things the command never guesses:

• Which column is the flag. A column called censored (or c) is read as 0 = failed, 1 = still running. Anything else, such as status, needs --censor-column status.
• Which way round it is. If your flag uses 1 for failed, add --c-invert rather than editing the file.

Every other tool works the same way: reliafy tools lists them, and reliafy <tool> <action> --help shows one action's arguments. More on the command line page.

4. The same from Python

The package you installed is also a Python client. Client() picks up the token reliafy login saved:

from reliafy import Client

rf = Client()
times = [412, 587, 655, 731, 790, 868, 944, 1010, 1105, 1190]
flags = [0, 0, 1, 0, 0, 1, 0, 0, 1, 0]      # 0 = failed, 1 = still running

result = rf.fit(times, distribution="weibull", censored=flags, unit="Hours",
                save_as="Bearing life")
print(result["url"])

rf.call(tool, action, ...) reaches every tool, with the same arguments the command line takes:

rf.call("strategy", "optimal_replacement", distribution_id="weibull",
        params=[{"name": "alpha", "value": 996}, {"name": "beta", "value": 3.8}],
        planned_cost=1000, unplanned_cost=8000)

A refused call raises reliafy.ReliafyError with Reliafy's reason in plain words. If you fit with SurPyval in your own code, reliafy.push(model, name=...) saves that fit, data and all. More on the Python client page, and complete scripts on Examples.

5. Push fleet usage every night

A fleet forecast is only as good as each item's current use. Rather than typing meter readings in, let your maintenance system's nightly export send them. The CSV needs the item's name (as it appears in the forecast) and its meter reading, and can say when it was read:

name,current_use,read_at
Truck 01,5230,2026-10-21T22:00:00+10:00
Truck 02,4817,2026-10-21T22:00:00+10:00

Copy the fleet's ID from its page (⋯ → Copy ID), then send the file:

reliafy ingest usage meters.csv --fleet FLEET_ID

Reliafy updates each item, recomputes the forecast, estimates each item's usage rate from the change since its last reading, and checks the fleet's alerts. A bad row stops the whole file with a message, and nothing is changed.

To run it every night, add a line to the crontab of the user who ran reliafy login (crontab -e). This one sends the export at 2:15 am:

15 2 * * * /usr/local/bin/reliafy ingest usage /data/exports/meters.csv --fleet FLEET_ID >> $HOME/reliafy-ingest.log 2>&1

Use the full path to reliafy (which reliafy shows it), because cron runs with a short PATH. A token with only Push data is enough. Degradation readings (reliafy ingest measurements) and new failures for a dataset (reliafy ingest lives) work the same way; see Original routes and ingest.

Files and photos

reliafy upload sends a CSV or Excel file as a dataset. reliafy import rbd brings in a diagram from BlockSim, Open-PSA, Galileo or RePyability JSON, and reliafy import excel reads an .xlsx workbook: a sheet as a dataset, an FMEA or RCM worksheet as an RCM study, or Reliafy's RBD template as a diagram. The details are on Files and photos; to import from the app instead, see import a diagram.

Where to go next

• The tools API lists every tool and action, live, with the OpenAPI spec for generating your own client.
• To work through the same tools in conversation, connect Claude (MCP).
• To pass a model to a colleague rather than a script, see share your work.