Agent Playground¶
Agent Playground is the interactive, mock-first replacement for Demo Lab. It shows requests, tool calls, policy decisions, loop detections, token totals, and the resulting Replay trace without requiring an API key.
Scenarios are deterministic fixtures. A run passes through the real Guard pipeline and is stored in the same SQLite database as ordinary agent traffic. The Playground polls the run API while a scenario is active and links directly to Replay for deeper inspection.
API¶
| Method | Path | Purpose |
|---|---|---|
GET |
/api/v1/playground/scenarios |
List scenarios |
POST |
/api/v1/playground/runs |
Start a run |
GET |
/api/v1/playground/runs/{id} |
Inspect a run |
An optional OpenAI-compatible upstream can still be configured for normal Guard traffic. Playground fixtures remain deterministic so screenshots, tests, and workshops do not depend on a paid service.