Reliafy

Open-source reliability engineering. Fit life distributions to failure data, build reliability block diagrams, and turn fitted models into maintenance decisions — in one self-hostable web app.

A FastAPI backend does the statistics with SurPyval and RePyability and serves a React single-page frontend, so the whole app runs on a single port.

Automate it: the ingestion API pushes meter readings, inspection measurements, and new failure data from scripts and cron jobs. Fit a model in a notebook and push it from SurPyval (pip install reliafy-client).

Questions, bug reports and suggestions are welcome as GitHub issues. What's shipped is on the What's new page.

Features

• Life-distribution modelling — fit Weibull, Exponential, Normal, Lognormal, Gamma, and proportional-hazards models to exact, censored, and truncated data. Probability plots on distribution-specific paper with confidence bounds, goodness-of-fit, and a survival calculator.
• Reliability block diagrams — wire components in series, parallel, k-out-of-n, and standby on a drag-and-drop canvas; compute system reliability, MTTF, importance measures, and minimal path/cut sets.
• Degradation & RUL — fit per-unit degradation paths to a failure threshold, then track in-service items and predict when each will cross it (remaining useful life with credible intervals).
• RCM studies — Function → Functional failure → Failure mode worksheets where every maintenance decision links to the analysis that justifies it, re-validated live (supported / contradicted / inconclusive).
• Maintenance strategy — rank candidate distributions against your data, compare two designs head-to-head, and find the cost-optimal preventive-replacement interval.
• Data (under Modelling) — upload a CSV once and reuse it across models; saved models reopen instantly.

Quickstart (self-hosted)

docker compose up --build

Open http://localhost:8000. That's it — no account or login (the instance is yours, in single-user mode), data persists in a Docker volume, and a few sample datasets, models, and an example RBD are pre-seeded so there's something to explore.

Configuration (self-host)

Variable	Default	Purpose
AUTH_DISABLED	(unset)	true = single-user mode: no login, every request is one fixed local user. The supported mode for self-hosting.
DEV_USER_ID	dev-user	The single user's id in single-user mode.
MONGODB_URI	(unset)	MongoDB connection string. docker-compose wires this to its bundled MongoDB. Unset/unreachable ⇒ automatic fallback to an in-memory simulator (data not persisted; GET /api/health reports which).
MONGODB_DB	reliafy	Database name.
SEED_SAMPLES	true	Seed shared sample datasets/models/RBDs on startup (idempotent).
PORT	8000	Port the server listens on.
ANTHROPIC_API_KEY / OPENAI_API_KEY	(unset)	Optional: set your own provider key to enable the built-in AI assistant on your instance (with AI_PROVIDER = anthropic (default) or openai, and optional AI_MODEL). Without a key the assistant is hidden.

The frontend build has one flag: VITE_AUTH_DISABLED=true (a Docker build arg in docker-compose) bakes single-user mode into the bundle — no login page and no marketing pages, straight into the app.

Local development

Backend

python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
AUTH_DISABLED=true uvicorn backend.main:app --port 8000 --reload

Frontend

Hot-reload dev server (proxies /api to the backend on :8000; frontend/.env.development already sets single-user mode):

cd frontend
npm install
npm run dev

Or a production build served by the backend on one port: npm run build, then open http://localhost:8000.

Tests

pip install pytest pillow-heif   # pillow-heif writes the tests' HEIC photos
pytest backend/tests

Usage

1. Upload a CSV.
2. Map your columns to SurPyval inputs.
3. Fit, inspect the probability plot, and save the model.

Column mapping

The mapping mirrors SurPyval's data model (surpyval.utils.xcnt_handler):

Field	Meaning
x	Observed values (exact / censored)
c	Censor flag per row: 0 observed, 1 right, -1 left, 2 interval
n	Count of observations per row (positive integer)
xl	Interval lower bound (used with xr)
xr	Interval upper bound (used with xl)
tl	Left truncation bound (-inf ⇒ untruncated)
tr	Right truncation bound (+inf ⇒ untruncated)

Rules enforced on both the frontend and backend:

• Use x or the xl/xr interval pair — never both.
• xl and xr must be supplied together.
• c, n, tl, tr are optional modifiers.

Left/interval-censored data is plotted with SurPyval's Turnbull estimator.

How persistence works

Datasets, models, and RBDs are MongoDB documents (uploaded CSV bytes are stored inline, content-addressed, on the dataset document). Saving a model stores the fit recipe plus cached results, so reopening is instant; SurPyval models can't be serialised, so covariate evaluation on a reopened proportional-hazards model re-fits on demand from the dataset + spec.

Reliafy Cloud

The hosted version at Reliafy Cloud adds what only makes sense hosted: user accounts (Firebase sign-in), a Pro plan, a metered AI assistant on prepaid credits, team workspaces (view free, edit with Pro), and view-only sharing of any analysis with other accounts. The cloud code lives in this same repository — GitLab-style, one codebase — and is dormant unless configured: without Firebase/Stripe/AI env vars, none of it runs. Multi-user env reference for operators:

Variable	Purpose
GOOGLE_CLOUD_PROJECT / FIREBASE_PROJECT_ID	Firebase project whose ID tokens are accepted (multi-user mode).
VITE_FIREBASE_*	Public Firebase web config baked into the frontend build (frontend/.env.production).
BILLING_ENABLED, STRIPE__API_KEY, STRIPE__WEBHOOK_SECRET, STRIPE__PRICE_ID	Plans, credit packs, and the Stripe webhook.
AI_PROVIDER, AI_MODEL, AI_MARKUP	The metered assistant's provider/model/markup.
PRO_MONTHLY_CREDIT_CENTS	AI credit included with each month of Pro (granted per paid invoice; default 1000).
ADMIN_EMAILS	Comma-separated operator emails with full access regardless of payment (no plan caps, AI not credit-checked).
PUSHOVER_TOKEN, PUSHOVER_USER	Operator push notifications (new signups, crashes) via Pushover. Unset → falls back to email, then to a log line.
PUSH_ERROR_ALERTS	Push an alert when the server hits an unexpected error (default on wherever Pushover is configured). Throttled: same fault once per 30 min, 8 per hour. Set false to keep a dev machine quiet.
SMTP_HOST, SMTP_PORT, SMTP_USER, SMTP_PASS, EMAIL_FROM	Outbound email for team invites and share notifications. Unset → every send is a logged no-op.

API (selected)

• GET /api/config — deployment capabilities {auth, ai, billing} (public).
• GET /api/health — liveness + storage mode (public).
• GET /api/distributions — {id, name, covariates} models that can be fit.
• POST /api/columns — multipart file; returns {columns, preview, n_rows}.
• POST /api/fit/{distribution} — multipart file plus any of the column fields x, c, n, xl, xr, tl, tr; returns parameters + probability-plot data.
• POST /api/models — fit and save; plus GET/PATCH/DELETE /api/models/{id} and POST /api/models/{id}/evaluate.
• GET|POST /api/datasets, GET|DELETE /api/datasets/{id}.
• GET|POST /api/rbds, GET|DELETE /api/rbds/{id}, POST /api/rbds/analyze, POST /api/rbds/validate.

License

Reliafy is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0) — see LICENSE. You're free to use, modify, and self-host it; if you offer a modified version as a network service, the AGPL requires you to publish your modifications. Contributions are welcome — see CONTRIBUTING.md.