Playground | Respan Docs

Build with the Respan MCP

We recommend installing the Respan MCP so your AI coding tool can work with your prompts, logs, and traces directly. Authenticate with your Respan API key from the API keys page:

$ claude mcp add \
>   --transport http \
>   --header "Authorization: Bearer YOUR_RESPAN_API_KEY" \
>   respan \
>   https://mcp.respan.ai/mcp

See the MCP docs for OAuth setup and other clients.

Use the Prompt Playground to build a conversation, configure one or more models, run the same input against each configuration, and compare the resulting content, latency, tokens, and cost.

Choose New or Legacy

Open the mode selector beside Playground:

New opens the current multi-column workspace described in this guide.
Legacy opens the previous Playground interface for workflows that still depend on its variant-based layout.

Respan Playground with the mode menu open, offering Legacy and New while a two-model comparison workspace is visible. — Use New for the current comparison workspace or Legacy for the previous Playground interface.

Start with the context you need

You can open a blank Playground or carry an existing prompt or production record into it.

From the prompt editor

Enter the prompt’s variable values, then select Playground in the top bar. This is useful when you want to test an uncommitted change or compare a prompt against another model before updating the prompt library.

Respan prompt editor with the action for opening the current prompt in Playground. — Open a prompt in Playground with its current messages and variables.

From production logs

Select Open in Playground on a log entry to reproduce its recorded input in an editable workspace. Remove customer identifiers, secrets, and sensitive prompt content before sharing or saving the reproduction.

Respan production log with the Open in Playground action available. — Send a recorded production interaction to Playground for reproduction.

Build the conversation

The New workspace represents the request as ordered message blocks:

System defines the behavior, constraints, and response style that apply to the run.
User contains the input you want the model to answer.
Assistant can supply prior assistant turns or an expected continuation.

Use the role control below the messages to add another User or Assistant turn. Keep the same conversation across compared models so the model or configuration is the only variable you change.

When testing safety or policy behavior, include the governing system context as well as the triggering user message. A user message by itself does not reproduce the full interaction.

Configure the run

Use the right panel before selecting Run:

Area	Use it to
Settings	Choose the model for each column and open its configuration.
Variables	Supply values referenced by the selected prompt.
Model settings	Adjust supported model-specific generation controls.
Parameter overrides	Override request parameters for this run.
Tools	Add tool definitions the model can call.
MCP servers	Give the run access to configured MCP tools.
Presets	Apply a reusable group of Playground settings.

Change one variable at a time when you need an attributable comparison. For example, keep the prompt and parameters fixed while comparing two models, or keep the model fixed while comparing two parameter sets.

Compare models side by side

Enable Compare

Select Compare in the header. The workspace adds Model A and Model B configurations and aligns their outputs in separate columns.

Choose each model configuration

Select the model for each column, then adjust settings or overrides only where the comparison requires a difference.

Run the same conversation

Select Run. Read the responses side by side, then compare the token, word, character, cost, and latency values displayed with each completed response.

Respan New Playground with System and User messages, side-by-side Assistant outputs, Model A and Model B selectors, and settings for parameters, tools, MCP servers, and presets. — The New Playground compares two model configurations against the same conversation and reports response metrics beside each output.

Save the workspace

Select Save in the header when you want to preserve the current Playground configuration. If you entered from a managed prompt, return to the prompt editor and commit the tested prompt change there so applications can reference the saved version.

Use the Legacy workflow

Legacy Playground uses a variant count and a Variants tab rather than the New workspace’s Model A and Model B columns. Set the number of variants, run the prompt, and compare the generated responses in that tab.

Respan Legacy Playground with the variant-count control used to generate multiple responses. — Legacy Playground generates the selected number of variants for comparison in its Variants tab.