Skip to main content
Conversations are the individual interactions that Coval evaluates. They come from two sources:

Simulated conversations

Conversations Coval creates by having a persona interact with your agent. Use them to test scenarios before they reach production.

Uploaded conversations

Production conversations submitted through the app, API, CLI, SDK, or an integration. Use them to measure what happened with real users.
Both sources produce transcripts, audio when available, traces, and metric results that you can inspect in Human Review, dashboards, reports, and Trace Search.

​
Conversations and runs

A run is a batch of simulated conversations launched with a shared configuration. One run can contain many simulated conversations, often one for each test case and persona combination. Uploaded conversations are submitted individually and do not use runs as their primary organization. The API and data model continue to use technical names such as simulation_id, conversation_id, and SimulationOutput. These identifiers describe stored resources and remain valid even though the product groups both sources under Conversations.

​
Choose a source

Use conversation_source when a filter or API contract needs to distinguish the sources. Its values are SIMULATED and UPLOADED; filters that span both may also accept ALL.

​
In the app

Open Conversations > Simulated to launch and inspect simulated conversations. Open Conversations > Uploaded to submit and inspect production conversations. Human Review and Metrics operate across both sources. Existing links under /runs, /conversations, and /monitoring continue to redirect to the corresponding Conversations page.

​
Next steps