# Quick start for vibe coders

This guide will show you how to build your first agent with the Toolhouse Agent Studio.

## Step 1: Describe your agent

First, [navigate to Agent Studio.](https://studio.toolhouse.ai/) Then, press **Create new Agent.**

You will find a chatbot like interface like this.

<figure><img src="https://3896932452-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fio16js3ONksvIYTf7YUl%2Fuploads%2FoYJKNGzm2WFs5LVMPLDD%2Fimage.png?alt=media&#x26;token=2aff6809-4960-486a-865a-e4fba82c7eec" alt=""><figcaption></figcaption></figure>

In the chat box you can **describe** what you want the agent to do, Toolhouse will **write the agent for you.** Make sure you are **specific** with your demands.

#### Channel your inner prompt engineer

Agent Studio helps you write the right prompt for your agents. You will just need to start in your own words, adding any details you may feel relevant.

<figure><img src="https://3896932452-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fio16js3ONksvIYTf7YUl%2Fuploads%2FCkK6lCuNFIq6Naz2gtnj%2Fimage.png?alt=media&#x26;token=972bdad3-b3ae-4e77-ba13-b122a4870f35" alt=""><figcaption></figcaption></figure>

### Example of prompts (with breakdown)

<details>

<summary>Good prompt</summary>

{% code overflow="wrap" %}

```
Build an agent that searches X for positive, negative, and neutral mentions about my product. The agent should give me a nice report and ask it if I want the report sent via email.
```

{% endcode %}

**Why it’s good:**

* Specific task (search for sentiment mentions about a product)
* Clear outputs (report, sentiment analysis)
* Defined interaction (ask about emailing)
* Useful for automation or monitoring

</details>

<details>

<summary>Bad prompt</summary>

```
Build an agent that finds competitors helps me with my business.
```

**Why it’s bad:**

* **Too vague**: finding competitors is a good start, but being specific could be better (search x, what type of competitors are you looking for”
* **Undefined scope**: What kind of business help? Marketing, sales, operations?
* **No clear task or output**: There’s no goal, input, or expected result.
* Large Scope “ finding competitors” will bring inconsistent results due to the wide scope the agent will search

</details>

{% hint style="success" %}

#### No one was born a prompt-engineer on their first day

Remember: you can have a natural conversation with Agent Studio, and add as many details as you can, even after it generates your first prompt.

**Give your agent an exit strategy.** For example, you can tell Agent Studio "if the agent is searching the web, limit it to 5 results"

**Tell your agent what it should do or should not do.** For example, you can tell Agent Studio: "never show results from apartments.com"

**Give your agent a tone, a voice, a personality.** If you need, you can make the agent respond in a particular way. Try telling Agent Studio: "always make the agent ask a followup question"
{% endhint %}

## Step 2: Run your agent

Once you've prompted your agent, you can run it to see how it responds.

When you run the agent it will reply to your first **`message`** (which you can see at the bottom of the YAML file, usually after **`system_prompt`**).

The `message` is what the agent will use as instructions when it runs the first time you press <i class="fa-play">:play:</i> **Run**.

<figure><img src="https://3896932452-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fio16js3ONksvIYTf7YUl%2Fuploads%2FfIFNloMIitQ4DMXbFtlj%2Fimage.png?alt=media&#x26;token=669a75a4-db3d-450e-b7db-33478b75e3ab" alt=""><figcaption></figcaption></figure>

## Step 3: Publish your agent

If the agent you built is satisfactory, press deploy to make it **live**.&#x20;

If not, then you can continue chatting with the agent studio to refine it.

<figure><img src="https://3896932452-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fio16js3ONksvIYTf7YUl%2Fuploads%2FRzsqHqXiRnDG4jNomDzT%2Fimage.png?alt=media&#x26;token=e0a0dfe4-c3c1-460c-baa5-e4aec1523d54" alt=""><figcaption></figcaption></figure>

At any point you can click  <i class="fa-upload">:upload:</i> **Publish** to publish your agent. Publishing your agent will make it available for use, so you can call the agent directly or integrate it with other tools you're using.

You will find your published agents under the **Agents tab.**

<figure><img src="https://3896932452-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fio16js3ONksvIYTf7YUl%2Fuploads%2FUpwNtp8cBFYfCp9NKiVh%2FScreenshot%202025-07-21%20at%2010.05.27%E2%80%AFAM.png?alt=media&#x26;token=65d4df2a-7deb-4583-a3f9-37e63cd5c52d" alt=""><figcaption></figcaption></figure>

## Step 4: Integrating your agent

When your agent has been published you can start using it immediately.

### **Use your agent immediately**

Each agent comes with a built-in chat interface, hosted by Toolhouse. This chat allows you and your end users to interact with your agent without the need to build from scratch.

To use it, simply click <i class="fa-upload">:upload:</i> **Publish**, then click the Agent interface link:

<figure><img src="https://3896932452-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fio16js3ONksvIYTf7YUl%2Fuploads%2FiAGSCjwOJmCCsRhxDJUT%2FScreenshot%202025-07-21%20at%2010.23.10%E2%80%AFAM.png?alt=media&#x26;token=9eb900f8-9ee9-4fb1-82b5-eac6d23e35eb" alt=""><figcaption></figcaption></figure>

Once you click it you will be redirected to an Agent interface where you can interact with your agent, just like ChatGPT, except it's already connected to all the MCP servers Toolhouse offers by default.

### Schedule your Agent for later

Toolhouse allows you to schedule agents so they can run at certain intervals.

To do so, click <i class="fa-upload">:upload:</i> **Publish,** then can go to the **Schedule** tab.

<figure><img src="https://3896932452-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fio16js3ONksvIYTf7YUl%2Fuploads%2FTliq99hwcaanNY9fS9ro%2Fimage.png?alt=media&#x26;token=430a6fa2-0382-4f98-9566-832f3d2392d8" alt=""><figcaption></figcaption></figure>

All you need to do from here is set a schedule for your agent to run and click **Create schedule.**

<figure><img src="https://3896932452-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fio16js3ONksvIYTf7YUl%2Fuploads%2FNw1ECuQpjh63fEz4Fw12%2Fimage.png?alt=media&#x26;token=3496e4b5-0156-47d7-acb6-d9c6fa272c83" alt=""><figcaption></figcaption></figure>

**Examples of agents you can schedule**

{% code overflow="wrap" %}

```
Find me trending AI events in SF and send an email to [insert your email here]
```

{% endcode %}

### **Vibe code your frontend**

Navigate to the **Lovable** or **Bolt** tabs under the Agent Deployment Success section, then press the clipboard icon to copy the prompt.

<figure><img src="https://3896932452-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fio16js3ONksvIYTf7YUl%2Fuploads%2FyuRGMBSuTiItjOxL05nt%2Fimage.png?alt=media&#x26;token=7a8b4374-b729-4e6a-afd4-27fc99f69166" alt=""><figcaption></figcaption></figure>

Go to [Lovable](https://lovable.dev/?ref=toolhouse) or [Bolt](https://bolt.new/?ref=toolhouse) and paste your prompt, then press Enter.

<figure><img src="https://3896932452-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fio16js3ONksvIYTf7YUl%2Fuploads%2FZXnNPOnJMSzb10v1QVwi%2Fimage.png?alt=media&#x26;token=6e75c961-e2fd-4642-879d-6f6a8a6dc4a8" alt=""><figcaption></figcaption></figure>

You will build an app that's automatically connected to Toolhouse. You can continue to build your project from there with Lovable, but your backend is now up and running!

## Get Help

You can find help from our[ AI support agent](https://help.toolhouse.ai/) live: our help chatbot can answer any questions.

Alternatively you can [join us on Discord ](https://discord.toolhouse.ai/)where the Toolhouse team and the community will help you out.
