Skip to Content

Playground

Interactive testing environment for AI models and agents.

Chat Interface

/dashboard/playground — test models with a conversational UI:

  • Model selector — choose from all available models (GPT-4o, Claude Sonnet, Gemini, etc.)
  • System prompt — configure agent behavior
  • Chat panel — send messages, view streaming responses
  • Configuration sidebar — temperature, max tokens, top-p, structured output schema

Features

Streaming Responses

Responses stream in real-time using the model service’s chat streaming API. Token-by-token display with typing indicator.

Process Timeline

Each chat turn renders an inline timeline of the agent’s thinking and tool-call steps, in the order they happened, alongside the streamed response — so you can see what the agent reasoned about and which tools it invoked without opening a separate log view.

The timeline is built from the same SSE stream as the chat response. One integration gotcha for anyone consuming this stream directly: tool identity is nested under metadata.tool/..., not at the top level of the event. A naive reader that looks for a top-level tool or tool_name field will miss it — always read tool identity from the event’s metadata object.

Structured Outputs

Enable JSON schema validation for structured responses:

  • Define output schema in the config panel
  • Model responses are validated against the schema
  • Invalid responses show validation errors

Model Comparison

Use the Feature Matrix to compare model capabilities before selecting in the playground.

MCP Playground

/dashboard/mcp/playground — test MCP tools directly:

  • Select any registered MCP tool
  • Fill input parameters from the tool schema
  • Execute and inspect JSON-RPC request/response
  • View raw protocol messages
Home