Using the Scratchpad

The Scratchpad is a rapid prototyping environment for testing prompts, iterating on agent behavior, and drafting workflow chains before committing them to production workflows.

Prerequisites

• Cloud Getting Started completed
• At least one agent created

---

Overview

The Scratchpad provides:

Feature	Description
Canvas	Freeform workspace for organizing prompt experiments
Draft execution	Run prompts without creating formal Lenses
Prompt testing	Test prompt variations side-by-side
Version history	Track changes to your experiments
Context management	Attach files, URLs, and data to prompts

---

Getting started

1. Navigate to your AI Lenser profile → Scratchpad tab
2. The canvas opens with an empty workspace
3. Use the toolbar to add elements

---

Canvas interaction

Adding elements

Element	Purpose	How to add
Prompt block	A text prompt to execute	Click + on toolbar
Result block	Shows execution output	Auto-created after run
Note	Freeform text annotation	Right-click → Add Note
File attachment	Attach context files	Drag and drop onto canvas

Navigation

Action	Gesture
Pan	Click + drag on background
Zoom	Scroll wheel or pinch
Select	Click on element
Multi-select	Shift + click
Delete	Select + Backspace

---

Draft execution

Running a prompt

1. Create a Prompt block
2. Write your prompt text
3. Select an agent from the dropdown
4. Click Run
5. The output appears in a connected Result block

Comparing outputs

Run the same prompt with different:

• Models — compare GPT-4o vs Claude side-by-side
• Prompts — test variations of the same instruction
• Parameters — try different input values

---

Prompt testing workflow

Iteration cycle

1. Write prompt variant A
2. Run with Agent X → observe output
3. Modify prompt → variant B
4. Run with Agent X → compare
5. Try with Agent Y → cross-model comparison
6. Pick the best → promote to Lens

Promoting to a Lens

When you find a prompt that works well:

1. Click the Promote to Lens button on the Prompt block
2. The Lens creation dialog opens pre-filled with your prompt
3. Define parameters using [[param]] syntax
4. Set metadata and visibility
5. Click Publish

---

Context management

Attaching context

Add context to your prompts:

1. Files — drag documents, code files, or data files onto the canvas
2. URLs — paste a URL to fetch and attach its content
3. Previous outputs — connect a Result block as context for the next prompt

Workbench context

The Scratchpad maintains a shared context workspace:

Context type	Persistence	Scope
Session context	Current session	This scratchpad only
Pinned context	Permanent	Across sessions
Agent memory	Agent-level	All runs for this agent

---

Running automation chains

Chain multiple prompts together in the Scratchpad:

1. Create Prompt A
2. Run it → Result A appears
3. Create Prompt B that references Result A:

   Based on the following analysis:
   {{result_a}}
   
   Now provide recommendations...

4. Run Prompt B → Result B appears
5. Continue chaining as needed

When the chain is validated, promote the entire sequence to a formal Workflow.

---

Version control

Auto-save

The Scratchpad auto-saves every 30 seconds and on significant changes.

Viewing history

1. Click the History icon in the toolbar
2. Browse timestamped snapshots
3. Click any snapshot to preview
4. Click Restore to revert to that state

Comparing versions

Select two history entries to see a diff of:

• Added/removed prompt blocks
• Changed prompt text
• Different execution results

---

Best practices

1. Use descriptive labels — name your prompt blocks clearly
2. Keep experiments focused — one hypothesis per scratchpad session
3. Document findings — use Note blocks to record observations
4. Promote winners — move validated prompts to Lenses promptly
5. Clean up — archive completed experiments

---

Next steps

• Building Workflows — formalize your scratchpad chains
• Creating AI Agents — configure agents for better results
• Team Collaboration — share scratchpads with teammates