Autonom documentation
Let your agent drive an iOS Simulator or an Android emulator, and bring back proof of what it saw.
Quick start
Four steps in a terminal, then your agent takes over.
- Get the code.
Clone the repository and step into it.
git clone https://github.com/aiatsuk/autonomcd autonom - Install it.
A checklist asks which device tools to add. Tick your agents too, and each one gets the skills.
./install.sh - Ask what this machine can do.
Green means installed, not proven. Anything missing comes with the exact fix.
autonom doctor - Take the tour.
Start a new agent session so it loads the skills. Then take the guided walk on your device.
autonom tour
Topics
Six groups, in the order you are likely to need them.
Start
Install Autonom, run a first check and learn how your agent owns a device.
- New to Autonom? Start with your first checkPoint your agent at a Simulator or emulator and get back screenshots and a journal.Start here
- InstallPut the command line on your PATH and give your agent the skills.
- Sessions and targetsOwn one device at a time and read every step back.
- Skills and routingSee which skills your agent loads for Flutter, native Android or iOS.
Drive
Read the screen and act on it by what each element says.
Observe
Keep what proves a change, see what the app sent, and hand a run to your agent.
Automate
Save a check once, run it in CI on every change, and pre-test a ticket before QA.
- FlowsWrite a check as a YAML file and run it again.
- App memoryKeep notes and runbooks for each app.
- Ticket pre-testCheck one ticket and its diff on a device, and hand QA a verified pack.
- ReplayRebuild the state of a recorded run at the step you choose.
- CI campaignsRun flows as one gated campaign, across runners, with exit codes.
Go further
Measure the app, and watch the device in your browser.
Reference
Look up an error code, what consent covers, and what a target can do.
How Autonom works
Skills and one command line. Your agent loads skills that say what to do, and runs
autonom, a command line that answers in JSON. No MCP server is needed.One explicit target. A session owns one device. With more than one ready, your agent names the one it means, or gets an error listing the candidates.
The evidence ladder. Each rung is stronger evidence, from code and tests to the screen, logs, performance and a before and after replay. Your agent climbs as far as the claim needs.