Prompt Lab

Overview

Prompt Lab is an interactive scratchpad for iterating on prompts without the overhead of creating an agent or building a workflow. You write a system prompt, type a user message, pick a model, and click Run — the response appears immediately. Change the prompt and run again. Repeat until the output quality meets your bar.

Once you have a prompt you are happy with, copy it directly into Agent Architect or into a node's system prompt field in the Workflow Editor.

Lab Interface

System Prompt Panel

The left panel is where you write the system prompt. This is the instruction that defines the model's role and behaviour. Changes here apply to every subsequent Run in the current session.

Conversation Panel

The right panel shows the conversation history. Each Run appends the user message and the model's response. You can send follow-up messages to test multi-turn behaviour, or reset the conversation to start fresh with the same system prompt.

Model Selector

The model selector at the top lets you compare the same prompt across different models. Run on Claude Sonnet, switch to TinyLlama, run again, and compare the outputs side by side in the history. This is the fastest way to choose the right model for cost/quality trade-offs.

Saving Prompts

Click Save Prompt to store a (name, system-prompt, model) tuple in the Prompt Lab library. Saved prompts appear in the library panel on the left — click any to load it into the editor instantly. This is useful for building a team library of vetted, reusable prompts.