Run agents safely without side-effects, supply realistic test inputs, step through execution node by node, and build a repeatable test suite before every deployment.
What sandbox mode does
Sandbox mode runs your agent with full logic — real LLM calls, real data-lookups — but with external side-effects intercepted. By default, outbound HTTP requests to third-party APIs return a configurable mock response instead of actually calling the external service, and write operations to databases, email services, and CRM systems are recorded but not committed. This lets you iterate quickly on your agent's logic without accumulating test records in your production systems, exhausting API quotas, or sending accidental emails. You can selectively disable sandboxing for specific nodes if you want to test their real responses.
Supplying test inputs
When you click Run in Sandbox, Cotonity presents an input form based on your agent's trigger schema. Fill in realistic test values that cover the scenarios you want to validate. You can save named test cases — for example, 'Happy path', 'Missing email field', 'Large account with 500 contacts' — and replay them at any time from the Test Cases panel. For webhook triggers, you can paste a real payload captured from your external system using the Capture Webhook feature, which temporarily routes one live webhook request to the sandbox instead of your production agent.
Inspecting the run step by step
After a sandbox run completes, the Run Inspector shows every node that executed, its inputs and outputs, the time it took, and the LLM token count for any AI nodes. Click any node to expand its detail view. If an LLM node made tool calls, you can see exactly which tools it called, with what arguments, and what they returned. Use the Step-through mode (the bug icon in the toolbar) to pause execution after each node and inspect the canvas state in real time before allowing the next step to run. Step-through mode is especially useful for debugging complex branching workflows where the control flow is not obvious from the canvas alone.
Building a test suite
Before deploying any significant agent, build a test suite of at least five to ten test cases covering: the happy path with typical inputs, edge cases with empty or null fields, inputs that should trigger the error path, inputs that exercise each branch in your conditional logic, and at least one large or complex input that tests performance. Run the full test suite before each deployment using the Run All Tests button in the Test Cases panel. Cotonity records the pass/fail status of each test case over time, giving you a test history that shows whether a recent change broke a previously passing scenario.