Backends

Each backend is a consumer of an EventStream — the timing engine never performs I/O itself (see Home → Core design decision). Every optional dependency is lazily imported inside the function that needs it, so import humaninput never requires any of them.

backend	dependency	does
terminal	none	replays typing into stdout with real timing (the default)
json	none	dumps the raw event stream
svg	none	animated SVG with a blinking cursor, no JS
css_keyframes	none	a CSS @keyframes typing animation
asciinema	none	writes a .cast v2 file
matplotlib	matplotlib	plots trajectories and interval histograms
pynput	pynput	drives the real OS keyboard/mouse
browser	none (BYO Playwright Page)	raw CDP key/mouse dispatch into a live page
touch	none (BYO Playwright Page)	raw CDP touch dispatch into a live page

terminal

from humaninput.backends import terminal

terminal.play(stream, speed=1.0, out=None, width=None)

Writes keydown characters at the timestamps in the stream, scaled by 1/speed. Row-aware redraw via ANSI — once the buffer wraps onto multiple terminal rows, a naive clear-line-and-redraw only fixes the last row, so this repaints all wrapped rows. width defaults to the real terminal width.

json

from humaninput.backends import json as json_backend

json_backend.to_json(stream, indent=2) -> str
json_backend.write(stream, path_or_file, indent=2)

Dumps the raw event stream (EventStream.to_dicts()) — the format every other backend, and humaninput fit's round-trip tests, treat as the canonical interchange shape.

svg

from humaninput.backends import svg

svg.render(stream, **kwargs) -> str
svg.write(stream, path, **kwargs)

Animated SVG with a blinking cursor and no JavaScript — safe to drop straight into a README or any Markdown renderer that allows inline SVG (this is what generates the animation at the top of the README).

css_keyframes

from humaninput.backends import css_keyframes

css_keyframes.render(stream, **kwargs) -> str

A CSS @keyframes typing animation for embedding directly in a web page's stylesheet.

asciinema

from humaninput.backends import asciinema

asciinema.write(stream, path, **kwargs)

Writes an asciinema .cast v2 file — playable with the asciinema CLI or embeddable via asciinema-player.

matplotlib

from humaninput.backends import matplotlib as mpl_backend

mpl_backend.plot_trajectory(stream, show=True, ax=None)
mpl_backend.plot_interval_histogram(stream, profile=None, median_ms=None, show=True, ax=None)

Used for the validation figures in the README and for eyeballing a generated mouse trajectory or interval distribution during profile tuning.

pynput

from humaninput.backends import pynput as pynput_backend

pynput_backend.play_keys(stream)   # KeyEvent stream
pynput_backend.play_mouse(stream)  # MouseEvent stream

Drives the real OS keyboard/mouse. pip install humaninput[pynput]. Prints a warning on first use of either function per process. Implements a fail-safe modeled on pyautogui's: if the real cursor is found at a screen corner when a mouse event is about to be dispatched, playback aborts immediately (RuntimeError).

This is what humaninput/tools/automate.py and shadow_writer.py are built on — see Tools.

browser

from humaninput.backends import browser

browser.play_keys(stream, page)   # KeyEvent stream
browser.play_mouse(stream, page)  # MouseEvent stream

page is a live Playwright Page (sync API) that the caller already created — this module never imports playwright itself, it just calls page.context.new_cdp_session(page) and sends raw CDP commands (Input.dispatchKeyEvent, Input.dispatchMouseEvent). This is deliberate: Playwright's own page.type()/page.click() have their own internal pacing, which would throw away this library's timing model. Dispatching at the CDP layer preserves it.

click/dblclick marker events in a MouseEvent stream are skipped — the down/up pair already dispatched is what a real click is made of.

from playwright.sync_api import sync_playwright
from humaninput import Typist, profiles
from humaninput.backends import browser

with sync_playwright() as p:
    page = p.chromium.launch().new_page()
    page.goto("https://example.com")
    stream = Typist(profile=profiles.load("touch_typist"), seed=1).type("hello")
    browser.play_keys(stream, page)

touch

from humaninput.backends import touch

touch.play_touch(stream, page)  # TouchEvent stream

Same no-import pattern as browser, dispatching Input.dispatchTouchEvent. tap marker events are skipped — touchstart/touchend are the real input. Pairs with humaninput.touch.Touch (tap, drag) and the mobile_thumbs profile's [touch] section — see The model → Touch.

from humaninput.touch import Touch
from humaninput.backends import touch as touch_backend

t = Touch(profile=profiles.load("mobile_thumbs"), seed=1)
touch_backend.play_touch(t.tap(120, 640), page)