Your first check

Install Autonom, point your agent at a Simulator or emulator, and get back a screenshot and a journal of what it did.

Try it autonom tour

Needs a Mac with Xcode, or the Android SDK · git, Python 3.11 and Node 20.11 or later · Claude Code, Codex, Grok or another agent that loads skills

Where Autonom runs

Pick one target for this check. The same commands work on both.

Run your first check

Steps 1 to 5 happen in your terminal. From step 6 your agent does the work.

  1. Install Autonom

    Clone the repository, step into it and run the installer. Tick your agent in the checklist so it gets the skills. Install covers every option.

    git clone https://github.com/aiatsuk/autonom cd autonom ./install.sh

  2. Ask what this machine can do

    Each capability says whether it is ready, and anything missing comes with the command that installs it.

    autonom doctor

  3. Take the tour

    Start a new agent session so it loads the skills. Then read the tour: what Autonom has, the usual workflow and the targets on this machine.

    autonom tour

  4. Have a target ready

    iOS Simulator. List the Simulators Xcode has. Your agent boots the one it picks when it starts its session.

    xcrun simctl list devices available

    Android emulator. Start one and wait for it to boot. Swap in an AVD name from autonom devices.

    autonom devices boot --avd Pixel_9

  5. Watch the tour walk

    It walks three screens into Settings, takes a screenshot after each step and writes an HTML report. It boots a target if none is running.

    autonom tour --run

    Open the file named in report_html in your browser: each step with its screenshot, the UI hierarchy and the device log.

    { "ok": true, "mode": "run", … "run": { "status": "passed", … "report_html": "…/flows/fr_…/report.html", … } }

  6. Give your agent one sentence

    In the agent session from step 3, say what to check. Swap in a button and a screen from your app.

    Check that tapping Sign in with Email opens the sign-in screen, bring back before and after screenshots, and leave the session open.

    Your agent starts a session on one target, reads the screen, taps by label as Read and drive the screen shows, and takes the screenshots.

  7. Read what it did

    The journal lists every command your agent ran and every note it left, in order.

    autonom journal
    { "ok": true, "count": 5, … "verb": "screenshot", … "verb": "ui tap", … "text": "Sign in with Email opens the email sign-in screen" … "journal": "…/sessions/s_…/journal.ndjson" }

    Here the agent started the session, took the before screenshot, tapped, took the after screenshot and left a note: five entries.

  8. Close the session

    Stopping frees the target for the next check. The session folder and its journal stay on disk.

    autonom session stop

What done looks like.

  • Journal: journal.ndjson in the session folder, one entry per command or note.
  • Evidence: a before and an after screenshot of the screen you named.
  • Report: report.html from the tour walk, one screenshot per step.
Reproduce these outputs

The journal output comes from Autonom's fake Android device, run as below from the root of an Autonom d0f7211 checkout in bash or zsh. Ids, times and paths differ on every run. The tour output needs a real device and is an example.

export AUTONOM_HOME=/tmp/first-check/.autonom AUTONOM_FAKE_STATE=/tmp/first-check/fake-state.json rm -rf /tmp/first-check && mkdir -p /tmp/first-check echo "{\"ui_dump\": \"$PWD/tests/fixtures/woolbox/android_sign_in.xml\"}" > $AUTONOM_FAKE_STATE autonom() { python3 scripts/autonom.py --adb tests/fakes/fake_adb.py "$@"; } autonom session start --serial emulator-5554 --app-id com.example.app autonom screenshot --label before autonom ui tap --desc "Sign in with Email" --mode exact autonom screenshot --label after autonom note add "Sign in with Email opens the email sign-in screen" autonom journal autonom session stop

Next steps

Your agent can now run a check and bring back what it saw.