Pick your runtime. One line.

Drop in the binary.

Single static binary, ~8 MB, no runtime dependencies. macOS (arm64, x64), Linux (amd64, arm64), Windows (amd64) native. Also available as a Docker image.

Heads-up: macOS binaries aren't Apple-notarized yet (v0.1.x) — Gatekeeper will warn on the curl path. Use Homebrew to bypass, or run xattr -d com.apple.quarantine after install. Windows SmartScreen may warn on first run; click "Run anyway".

Check the version + signalling endpoint.

Confirms the binary is on your PATH and that it can reach the signalling service. fernsicht doctor runs a deeper set of checks (DNS, TLS, TCP, proxy detection).

→ fernsicht 0.1.x  (signalling: https://signal.fernsicht.space, reachable)

Put it in front of any command.

Anything that prints progress becomes observable. fernsicht run -- auto-detects common progress formats (tqdm, bracketed [N/M], N of M, bare percent, step/epoch/iter keywords). For anything custom, emit a __fernsicht__ magic-prefix line — see fernsicht magic.

→ https://app.fernsicht.space/#room=abc12345&role=viewer
  share this link — viewers open in any browser, no install

Flags worth knowing.

Sensible defaults. Reach for flags when you want a custom label, a cap on viewers, or a self-hosted signalling server. Full reference: fernsicht run --help.

Pure-Python package.

Pre-release — the wheel is not yet on PyPI. Install directly from GitHub for now; a PyPI release is coming with the next minor version.

Runtime deps: aiortc (WebRTC) and aiohttp. Python 3.9+ on any OS with a modern TLS stack.

Drop-in replacement for any iterable.

Reports progress, rate, and ETA. Works with any sized iterator — lists, generators, DataLoaders, ranges. URL prints once on first iteration.

from fernsicht import blick

for epoch in blick(range(1200), desc="Training model"):
    train_one_epoch(model, epoch)

→ https://app.fernsicht.space/#room=abc12345&role=viewer  (shown once on first iteration)

When you don't have a loop.

For streams, callbacks, multi-stage pipelines, or async producers — use the manual bar and call update() as progress accumulates. Close it explicitly (or with a with block).

from fernsicht import manual

with manual(total=n_files, desc="Preprocessing") as bar:
    for path in files:
        handle(path)
        bar.update(1)

Default caps at 8 concurrent viewers.

Same link works for everyone. The server caps rooms by default; pass max_viewers= to tighten (single-viewer) or raise (up to the server limit). No accounts, no logins, no roles.

from fernsicht import blick

for item in blick(items, desc="Team demo", max_viewers=1):
    process(item)

Install from GitHub.

Pre-release — CRAN submission comes with the 1.0 line. Pure R, no compiled code. Downloads a small signed helper binary on first use.

v0.1.x known issue: the lazy bridge-binary download points at a placeholder release until bridge/v0.1.0 ships. Track the releases page — once the bridge tag lands, blick() works end-to-end. CLI users can observe their R runs today via fernsicht run -- Rscript my-analysis.R.

One call over a list + a function.

Returns the collected results, with live progress streamed to any browser.

library(fernsicht)

results <- blick(items, process, label = "Sampling chains")

For custom loops and long simulations.

Open a session, tick once per unit of work. The session auto-closes when the calling scope exits. with_session() + tick() are the primitives behind blick().

library(fernsicht)

with_session(label = "Bootstrap", total = B, {
  for (b in 1:B) {
    resample(data)
    tick()
  }
})

Wrap any long render step.

Drop blick() around any loop in a code chunk — the URL prints once, viewers stay attached as the doc renders.

library(fernsicht)

res <- lapply(datasets, function(d) {
  blick(seq_len(nrow(d)), function(i) fit_row(d[i, ]),
        label = paste("fit —", d$name[1]))
})