> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kakiyo.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Understanding the Sandbox

> Test your AI agent before launching real campaigns. Learn how to access the Sandbox from the dashboard, simulate conversations with different prospect types, and optimize your prompts and offerings safely.

<Info>
  **This guide is for dashboard users.** Everything below happens inside the [Kakiyo dashboard](https://app.kakiyo.com). No code or API calls required.
</Info>

> The Sandbox is a safe testing environment where you can simulate conversations with your AI agent before it talks to real prospects. Use it to validate your prompts, offerings, and follow-up sequences without risking your reputation or wasting credits on poorly configured campaigns.

## How to Access the Sandbox

The Sandbox is campaign-specific. You need a campaign before you can use it.

1. In the sidebar, click **Campaigns**.
2. Click on the campaign you want to test.
3. Click the **Sandbox** tab (or the **Test in Sandbox** button).
4. The Sandbox chat interface opens.

<Note>
  If the Sandbox tab is not visible, make sure your campaign has an offering and prompt assigned. Both are required for the Sandbox to work. Your LinkedIn account does **not** need to be connected to use the Sandbox.
</Note>

***

## How the Sandbox Works

The Sandbox simulates a real LinkedIn conversation between your AI agent and a fictional prospect. Here's how it differs from real conversations:

|                                | Sandbox                    | Real Campaign |
| ------------------------------ | -------------------------- | ------------- |
| Messages sent to LinkedIn      | No                         | Yes           |
| Credits consumed               | Minimal (testing credits)  | Full credits  |
| Prospect is real               | No — you play the prospect | Yes           |
| AI uses your offering + prompt | Yes                        | Yes           |
| Follow-up messages work        | Yes                        | Yes           |

***

## Running a Sandbox Test

### Step 1: Start a New Conversation

1. In the Sandbox, click **New Conversation** (or start typing).
2. The AI agent sends its first message based on your First Message Prompt.
3. Review the message: is it personalized? Does it match the tone you configured? Is the CTA clear?

### Step 2: Play the Prospect

Now you role-play as the prospect to test how the AI handles different scenarios:

1. Type a reply as if you were a real prospect.
2. The AI responds based on your Context Prompt and Offering.
3. Continue the conversation to test specific scenarios.

### Step 3: Test Different Prospect Types

Run multiple conversations to cover these common scenarios:

| Prospect type      | What to test                                       | Example message                                                   |
| ------------------ | -------------------------------------------------- | ----------------------------------------------------------------- |
| **Interested**     | Does the AI qualify correctly and book a meeting?  | "That sounds interesting, tell me more about pricing."            |
| **Skeptical**      | Does the AI handle objections without being pushy? | "I'm not sure this is right for us. We already use a competitor." |
| **Busy**           | Does the AI respect the prospect's time?           | "Not a good time, maybe later."                                   |
| **Not interested** | Does the AI accept rejection gracefully?           | "No thanks, not interested."                                      |
| **Off-topic**      | Does the AI stay on track?                         | "Do you know any good restaurants nearby?"                        |
| **Aggressive**     | Does the AI remain professional?                   | "Stop messaging me, this is spam!"                                |

### Step 4: Test Follow-Up Messages

If you have follow-up messages configured (see [Follow-up Messages](/guides/core-features/follow-up-messages)):

1. In the Sandbox conversation, do not reply to the AI's first message.
2. Click the **Trigger Follow-up** button (or wait for the simulated delay).
3. The AI sends the first follow-up message.
4. Repeat to test the second and third follow-ups.
5. Verify that each follow-up adds new value and doesn't repeat the same content.

***

## What to Look For

When reviewing Sandbox conversations, check:

| Aspect                  | What to verify                                                                        |
| ----------------------- | ------------------------------------------------------------------------------------- |
| **Tone**                | Does the AI match the personality you configured? Professional, casual, friendly?     |
| **Product knowledge**   | Does the AI accurately describe your product based on the offering?                   |
| **Qualification**       | Does the AI correctly identify qualified vs. unqualified prospects?                   |
| **Boundaries**          | Does the AI respect the rules in your Context Prompt? (e.g., "never discuss pricing") |
| **Call to action**      | Does the AI propose a clear next step (book a call, share a link)?                    |
| **Handling objections** | Does the AI address concerns without being defensive or pushy?                        |
| **Message length**      | Are messages concise and readable, or too long and wall-of-text?                      |

***

## Adjusting Based on Sandbox Results

If the AI doesn't behave as expected, here's what to adjust:

| Problem                                   | Where to fix it                                                                                |
| ----------------------------------------- | ---------------------------------------------------------------------------------------------- |
| AI doesn't know enough about your product | Edit your **Offering** — go to **Offerings** in the sidebar, click the offering, edit          |
| AI tone is wrong                          | Edit the **Context Prompt** — go to **Prompts** in the sidebar, click the prompt, edit context |
| First message is off                      | Edit the **First Message Prompt** — go to **Prompts**, click the prompt, edit first message    |
| AI qualifies too easily or too strictly   | Adjust qualification criteria in the **Context Prompt**                                        |
| Follow-up messages are repetitive         | Edit follow-up templates in the campaign's **Follow-ups** tab                                  |
| AI gives incorrect information            | Add or correct facts in the **Offering**                                                       |

After making changes, run the Sandbox again to verify the improvement. Repeat until you're satisfied with the AI's behavior.

***

## Sandbox Best Practices

1. **Test before every campaign launch** — always run at least 3-5 Sandbox conversations before going live.
2. **Test after every prompt or offering change** — even small edits can change behavior.
3. **Test with different models** — if you're considering switching AI models, compare Sandbox outputs (see [How to Choose Your AI Model](/guides/inbox-conversations/how-to-choose-your-ai-model)).
4. **Save examples** — note down particularly good or bad Sandbox conversations to reference when optimizing.
5. **Involve your team** — have team members play the prospect for more realistic testing.

***

## Related Guides

* [Understanding the Prompts](/guides/core-features/understanding-the-prompts)
* [Understanding the Offerings](/guides/core-features/understanding-the-offerings)
* [Follow-up Messages](/guides/core-features/follow-up-messages)
* [How to Choose Your AI Model](/guides/inbox-conversations/how-to-choose-your-ai-model)