Python
scripting

Steps

Terms used

Instrument time is the time the operation would take on the instrument: mill() runs a pattern to completion at once and reports it. A challenge is a training task's hidden error, set up for a script instead of a person: the readings (focus, stage Z) differ from the truth by a hidden amount.

Rules

• While a person is doing a training task in the exhibit, scripts cannot change settings. The exhibit refuses, so recorded practice stays human.
• Challenge runs are never recorded or shared with the research dataset.
• Ending a challenge restores the instrument as it was before.
• In the IDE, the exhibit pauses while a script runs. Keep long milling loops on a desktop.
• Live mode listens on 127.0.0.1, on port 8792 (the one fibview-mcp uses). Run one or the other, or choose another port.

1. Install Node.js 20 or later (nodejs.org) and Python 3.9 or later.
2. Get the repository and install the package:

git clone https://github.com/Advanced-EM/FIB-view
cd FIB-view
pip install -e "tools/fibview-py[live,plot]"
python tools/fibview-py/examples/quickstart.py

3. fv.connect() starts the simulator in a Node.js process: no browser. If the package lives outside the repository, set FIBVIEW_ROOT to the repository folder.
4. To drive an open exhibit instead, use fv.connect("live"), then select Connect agent in the exhibit (Controls → Voice and agent) within two minutes.

The calls

Modes: se, inlens, bse, ebsd, tkd, eds, fib, mill, gis, sv, lam, mc. Specimens: chip, ni, sn, nmc, xs. Ions: Ga, Xe, Ar, N, O. info() lists them all.

Compare ion species

Mill the same 6 × 6 µm box, asked to reach 2 µm in silicon, with each species at ~4 nA (example mill_rates.py):

fib.set(specimen="chip", mode="mill")
for k, ion in enumerate(["Ga", "Xe", "Ar", "N", "O"]):
    fib.set(ion=ion, ion_current_pA=4000)
    r = fib.mill(shape="rect", w=6, h=6, depth=2,
                 x=-24 + 12 * k, y=0)
    print(ion, r["maxDepth"], r["Yv"], r["time_s"])

The gallium column's nearest aperture to 4 nA is 2.5 nA. Each pattern's dose comes from the silicon application file, so the depths are similar and the times show the speed. The oxide and copper of the chip mill at their own rates.

A challenge hides an error, as a training task does for a person. Your algorithm sees only the images and the readings; score() judges the true result afterwards. Example auto_focus_ion.py focuses the ion beam on tin spheres. Every image it takes sputters the sample, and the score counts that damage too. Focus first, then the stigmator, then focus again: astigmatism moves the best focus. With seed 7 it scores 100: 4 nm of blur on a 5.3 nm probe, after 37 frames that removed 4 nm of tin.

import numpy as np, fibview as fv
fib = fv.connect()
fib.challenge("fib-focus-stig", seed=7)

def sharpness():
    img = fib.acquire(steps=2)["ion_image"]
    gy, gx = np.gradient(img.astype(float))
    return np.mean(np.hypot(gx, gy)) / img.mean()

def search(key, lo, hi, n=7):
    best = None
    for v in np.linspace(lo, hi, n):
        fib.set(**{key: float(v)})
        s = sharpness()
        if best is None or s > best[1]:
            best = (v, s)
    fib.set(**{key: float(best[0])})
    return best

f, _ = search("ion_focus_um", -100, 100, n=9)
a, _ = search("ion_stig", -60, 60, n=9)
search("ion_stig", a - 10, a + 10)
f, _ = search("ion_focus_um", f - 25, f + 25)
search("ion_focus_um", f - 6, f + 6, n=5)
print(fib.score())     # blur vs the probe, imaging damage
print(fib.reveal())    # the hidden error
fib.end_challenge()

Save data for analysis

acquire() returns every array of the current mode and its metadata (pixel sizes, beams, patterns with their depth maps, the lamella's true geometry). a.save("run.npz") writes the same NumPy archive as the exhibit's Export button.

import numpy as np, fibview as fv
fib = fv.connect()
fib.set(specimen="nmc", mode="sv")   # slice & view
a = fib.acquire()
stack = a["stack"]                   # slices × depth × x
truth = a["truth_labels"]            # the true phases
print(a.calibration["fractions"])
a.save("cathode.npz")

Without Python

node tools/headless.mjs --mode mill --spec chip \
     --ion Xe --mill rcs:16x8x6 --out xsec.npz

Regression tests

Challenges take a seed, so the same hidden error comes back every run: an automation routine can be tested in CI:

# tests/test_autofocus.py
import fibview as fv

def test_ion_autofocus():
    with fv.connect("headless") as fib:
        fib.challenge("fib-focus-stig", seed=7)
        my_autofocus(fib)
        s = fib.score()
        assert s["success"], s

The package's own tests: pytest tools/fibview-py/tests.

For AI agents

The same instrument is available to AI agents through the MCP server tools/fibview-mcp: headless simulate, mill and ion_physics tools, and live control of an open exhibit.