everbar

A progress bar that works everywhere — terminal, Jupyter, JupyterLab, VS Code notebooks, Google Colab, Marimo, Pyodide, and CI logs. One API, the right backend per environment.

Status: 0.2.0 — alpha. API may shift.

Install

pip install everbar             # core only; uses text fallback if nothing else is installed
pip install "everbar[tqdm]"     # terminal + Jupyter via tqdm
pip install "everbar[notebook]" # tqdm + ipywidgets for notebook front-ends
pip install "everbar[all]"      # everything (tqdm, rich, ipywidgets, marimo)

Use

from everbar import Progress

for x in Progress(items, desc="Loading"):
    work(x)

with Progress(total=100, desc="Steps") as bar:
    for _ in range(100):
        do_step()
        bar.update(1)

Other options:

Progress(items, unit="files")   # label what's being counted
Progress(items, disable=True)   # render nothing (e.g. behind a quiet flag)

Live metrics with set_postfix

Show a live key/value suffix next to the bar — useful in training loops:

with Progress(total=epochs, desc="Training") as bar:
    for epoch in range(epochs):
        loss, acc = train_one_epoch()
        bar.set_postfix(loss=loss, acc=acc)
        bar.update(1)

Calling set_postfix again replaces the previous suffix. Floats are formatted compactly (e.g. loss=0.424, acc=0.91).

Signal failure with fail()

Mark the bar as failing without stopping it — useful when one task in a batch errors but the overall job continues:

with Progress(total=len(jobs), desc="Batch") as bar:
    for job in jobs:
        if not run(job):
            bar.fail()
        bar.update(1)

Rendering is backend-specific: red bar in tqdm, FAIL marker in Rich, [failing] log lines in non-TTY mode, a red badge in Marimo. The state is sticky.

Overrides

Progress(items, backend="terminal")        # per-call

EVERBAR_BACKEND=terminal python script.py  # env var

import everbar
everbar.set_default_backend("terminal")    # module-wide

Precedence: the backend= argument wins, then EVERBAR_BACKEND, then set_default_backend, then auto-detection.

Unknown backend names raise ValueError (EVERBAR_BACKEND warns and is ignored instead). If you request a backend in code (backend= or set_default_backend) and its dependency isn't installed, you get an ImportError; if EVERBAR_BACKEND names it, everbar warns and uses the text fallback instead — deploy-time configuration never crashes a script. Auto-detection falls back silently.

Extra keyword arguments to Progress are forwarded to the selected backend, so they are environment-specific by nature (tqdm's colour, Rich's console, the fallback's min_interval). Stick to the named parameters in code that must run everywhere.

How it picks a backend

everbar.detect_environment() returns one of: marimo, colab, kaggle, vscode_notebook, jupyter, spyder, databricks, pyodide, ipython_terminal, terminal, non_tty. Each maps to a backend, with graceful fallback to a log-line text mode when nothing better is available.