Autonomous Exploration for Test Cloud

Autonomous Exploration in Test Manager for having an AI agent explore an application against a requirement, charter, or user story, without a pre-defined test script.

Autonomous Exploration is a capability in Test Manager where an agent simultaneously learns about the application, designs tests, and executes them in real time, with no pre-defined test script. The agent explores the application like a curious user, observes behavior, and adapts dynamically. Autonomous Exploration is guided by a charter, requirement, or user story, and runs as an unattended automation.

Important:

Autonomous Exploration is available in Test Manager (Commercial delivery) only when delivered via Test Cloud.

Key features

• Discovery-driven: focuses on finding unknown issues, edge cases, and unexpected behavior, including complex, BDD-style (Given/When/Then) requirements, not just simple flows.
• Reasoning-driven: an artificial intelligence (AI) model interprets application state and decides the next action to take.
• Time-boxed sessions: a duration is set by the user, for example 10 minutes. The displayed run duration reflects only the exploration time-box, and is approximate, since a run can slightly exceed the box if it is still mid-action when time runs out.
• Execution scope: runs on a cloud robot or an attached robot (serverless robots are not supported), with only one exploration allowed per requirement at a time, and exploration confined to a single website or application.
• Custom prompts: a custom prompt can optionally be added through Project Settings > Prompt Library, used as User Instructions. There is no built-in charter prompt.
• Comprehensive reporting: a structured report of findings is generated after the exploration completes.

Human-in-the-Loop for Autonomous Exploration

Live streaming is opt-in. When enabled, logs poll approximately every 10 seconds, with the latest entry appearing at the top and the view auto-scrolling. Closing the side panel does not stop the exploration itself, only the live view; reopening it reconnects and restores state.

Human-in-the-Loop turns every exploration into a supervised session, with real-time control available from the side-panel console without leaving the panel or restarting the job:

• Pause and guide: the agent can be stopped mid-scenario, with a guidance message added directly in the console.
• Skip scenarios: anything not relevant to the current run can be skipped, from the scenario row.
• Resume or stop: the exploration can pick up where it left off, or stop gracefully; the agent still generates its report over everything explored so far.
• Pause does not cost time: the timebox countdown freezes while a response is being written. If not resumed manually, the agent auto-resumes after 5 minutes, so runs never silently stall.
• Flexible streaming: with live streaming on, the panel can expand full-screen to show the browser alongside the controls; with it off, the guidance chat appears inline as soon as the job is running.
• Secure by design: every command is authorized against the user who started the exploration, so only that user can control the run.

Known limitations

• Autonomous Exploration uses the project's default Orchestrator folder. No folder override is available.
• If no eligible robot is available, a run stays in Pending state rather than failing outright.
• Triggering exploration from Autopilot Chat, and Jira-integrated actionable findings, are not available.
• No secure credential-injection mechanism is available. Real credentials should not be entered in plain text in a requirement.

For step-by-step instructions, see Running Autonomous Exploration. For details on reviewing a run's results, see Reviewing Autonomous Exploration results.