github.com/elliotgao2/handsets @v0.1.38

Drive Android from the shell, in milliseconds. One jar; no app, no root.

$ hs use                              # connect, start the on-device daemon
daemon up on tcp:9008

$ hs ui                               # flat table of tappable nodes
fill  EditText  "Email"     #email     540,540
fill  EditText  "Password"  #password  540,640  [password]
tap   Button    "Continue"  #continue  540,860

$ hs tap  "Continue"                  # text-lookup tap
$ hs wait "Welcome"                   # block on success text

ui → label → tap → wait. Pipe hs ui into a model, take the label, hand it back. Raw screenshot defaults to 768px-long-edge JPEG, the fast agent path; hs see out.jpg saves a native-resolution export unless you pass --size 768. Use WebP for compact lossy exports and PNG only for debug/lossless files.

Reproduce with hs bench -n 50; methodology in docs/benchmark.md. uiautomator2 and Appium ship things Handsets doesn't — recorders, pytest plugins, HTML reports, iOS support. Handsets is the lean CLI for tap-heavy work where per-call cost matters.

hs tui — drive any Android app from your keyboard

A live, video-stream terminal UI of the device's interactive elements. A background watcher polls the accessibility tree at ~10 fps, so the list on screen is the live state of the device — no manual refresh, no waiting for idle, no race between tap and re-dump.

• One screen for the whole flow. ↑↓ / jk to move, Enter to tap a Button or open an input modal for an EditText, PgDn/PgUp or Shift+J/K to swipe, ← / Esc for BACK, q to quit. No more copy-paste between hs ui and hs tap "…".
• What you see is what you'd script. Rows render the same verb-led table as hs ui -i — fill EditText "Email" #email, tap Button "Continue" #continue — so once the flow works in tui, you already know the selectors for hs run automation.
• Fast. Actions are fire-and-forget over a dedicated socket; the watcher picks up the new screen on its next 100ms tick. Mash keys without waiting.
• Works on animated pages. No wait_for_idle blocking — even constantly animating screens render smoothly at 10–20 fps.

Open it with hs tui after hs use. Built as a sibling crate (handsets-tui, ratatui + crossterm) so the core hs binary stays zero-dep.

Selectors

CSS-like, Playwright-inspired:

hs tap   'EditText[hint~=Email]'             --visible --unique
hs fill  'EditText[hint~=Password]'          'hunter2'
hs tap   'Button:has-text("Sign in")'        --visible --unique --timeout 5s
hs wait  "Dashboard"                         --timeout 15s

Vocabulary: [a=v] [a~=sub] [a^=pre] [a$=suf] · :visible :clickable :enabled :focused :checked · :has-text("x") :text-is("x") · :in(SEL) :below(SEL) :right-of(SEL) :near(SEL, PX). Comma is OR. More patterns in docs/cookbook.md.

How it works

host                                  device (shell UID via app_process)
─────                                  ─────────────────────────────────
  hs ─── adb forward ────► tcp:9008 ─► Server.java
                                         ├─ accessibility dump
                                         ├─ binder reflection (Pm/Am/Wm/Settings)
                                         └─ screenshots + state mirror

Length-prefixed binary frames over adb forward:

host → device:   [u32 BE len][ascii: verb [k=v ...]]
device → host:   [u32 BE len][bytes]              ok body
                 [u32 BE 0]                       end-of-stream
                 ERR:<CODE>:<detail>              failure

Full protocol: docs/wire.md. Reflection details and sharp edges: docs/architecture.md, docs/sharp-edges.md.

Install

curl -fsSL https://raw.githubusercontent.com/elliotgao2/handsets/main/install.sh | bash

Needs adb on $PATH. macOS and Linux. Pin a version with HANDSETS_VERSION=v0.1.26.

Python bindings: pip install handsets.

from handsets import Session
with Session() as d:
    d.tap("Continue")
    d.wait(text="Welcome", timeout="15s")

For other languages, drive hs --json as a subprocess and parse one JSON line per call.

Exit codes: 0 ok · 1 failure · 2 NOT_FOUND · 3 TIMEOUT · 4 AMBIGUOUS. Full structured error.code in hs --json output for the long tail.

hs --help is the verb table. Recipes in the Cookbook, long-form posts on the Blog.

Pre-1.0. CLI surface stable since v0.1.0. Wire protocol versioned via the info verb. MIT — see LICENSE.