Goals
A goal is the objective you hand the machine — "deploy the landing page to staging". Automata interprets it into a concrete, reviewable objective before any work begins.
Every run follows the same loop — Goal, Plan, Act, Verify, Recover — with a human in command at the gates that matter. Here is the whole model.
A goal is the objective you hand the machine — "deploy the landing page to staging". Automata interprets it into a concrete, reviewable objective before any work begins.
Each goal becomes a step plan with explicit approval gates. You see the whole route before it runs and can edit where humans must sign off.
Automata attaches to a real browser and a code runtime, taking concrete actions — clicking, filling, running scripts — instead of just describing them.
Results are asserted against the goal: status codes, screenshot diffs, output checks. Nothing is "done" until it is verified, and the evidence is kept as artifacts.
Approval gates pause a run at the moments that matter. You approve, reject or delegate — keeping a person in command of every consequential action.
When a step fails, armed recovery retries with an adjusted strategy or stops cleanly and raises an approval — so a run never silently goes off the rails.
Add the accounts and credentials for the task into the scoped vault. The run only ever sees what you authorize.
Describe the outcome in a sentence. Automata interprets it and proposes a plan with approval gates for you to review.
Adjust the gates, hit Start Run, and watch the timeline as the machine acts, verifies and recovers.
Every run leaves screenshots, checks and an audit trail you can browse and export from the dashboard.
Open the dashboard, write a goal, and follow it through plan, execution, verification and recovery — live.