Agent Playground & Interactive Testing

The Agent Playground provides an interactive execution testbed to debug autonomous agents, inspect tool invocations, and observe reasoning steps before deploying to production.

• Portal Page: /projects/[projectId]/playground

sequenceDiagram
    actor Dev as Developer
    participant PG as Agent Playground
    participant Agent as Autonomous Agent
    participant Tool as Bound Tool (e.g. DB Query)

    Dev->>PG: Send Prompt: "What was customer #102's last order?"
    PG->>Agent: Execute Agent Reasoning Loop
    Agent-->>PG: Generate Thought: "I need to call query_customer_orders"
    PG->>Tool: Execute Tool {"customerId": 102}
    Tool-->>PG: Return Tool Output {"orderId": "ord_99", "amount": "$45.00"}
    PG->>Agent: Feed Tool Output back to Model
    Agent-->>PG: Synthesize Final Response
    PG-->>Dev: Display Formatted Message + Tool Trace Ladder

Key Features #

1. Multi-Turn Chat History: Test conversation continuity and context memory.
2. Execution Step Inspector: View exact tool arguments, raw tool responses, and model thoughts.
3. Parameter Overrides: Tweak temperature, max tokens, and system instructions on the fly.
4. Token & Latency Accounting: Real-time display of tokens consumed during multi-step tool calls.

Next Steps #

• Define agent skills: Skills & Orchestration
• Connect MCP tools: MCP Servers & Transports