Agent Tools

Agent Playground

A dedicated UI for testing agents without creating workflows. Send prompts, switch providers, and inspect raw responses.

Using the Playground

The Playground is accessible from the Agents page. Select an agent, enter a message, and choose a provider/model override if desired. The playground sends the message through the same LLM adapter used by workflow steps, so results are representative of production execution.

Agent Selection

Pick any registered agent. The playground uses the agent's system instructions and capabilities.

Provider Override

Override the agent's default provider and model for quick experiments.

Raw Output

Inspect the full LLM response including token usage, finish reason, and raw content.

Playground API

The playground calls the agent run endpoint under the hood.

POST /api/agents/:id/run

{
  "prompt": "Explain quantum computing in one paragraph.",
  "provider": "openai",
  "model": "gpt-4o-mini"
}

Notes

• Playground execution is rate-limited alongside other expensive operations.
• Memory is not persisted during playground runs unless explicitly enabled for the agent.