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

# Create your own Runner

> Add one AI employee with a valuable job, schedule, sources, rules, and tool permissions.

You're adding one AI employee, not asking a generic assistant to "help with work."

Give the Runner one job. Make sure that job protects revenue, removes cost, gives back meaningful time, or keeps a customer promise from slipping.

## Write the job description first

| Part               | What to decide                                                                  |
| ------------------ | ------------------------------------------------------------------------------- |
| **Job**            | The one result this Runner should produce                                       |
| **Value**          | The time, cost, revenue, or customer promise attached to it                     |
| **Sources**        | The connected apps, files, records, or sites it should use                      |
| **Schedule**       | When the job runs and which timezone it follows                                 |
| **Stop condition** | When Runner should report that no action is needed and stop                     |
| **Review**         | What can happen directly and what must wait for an Action Card                  |
| **Human work**     | The judgment, relationship, creative choice, or final decision that stays yours |
| **Score**          | The number you'll check after the first few runs                                |

There are two ways to add the Runner:

1. Fill in the form yourself.
2. Ask Runner to help you design the job.

Use the second one when the job is still fuzzy. Use the form when you already know the brief and schedule.

<Warning>
  **A new Runner is active as soon as it's created.** Check every field the creation path shows before you confirm. Then open the new Runner and review its Project, model, and tool permissions before the first scheduled time arrives.
</Warning>

## Ask Runner to create it with you

<Steps>
  <Step title="Start the setup">
    Open **Runners**, click **New runner**, then choose **Ask Runner to create it**. Set the new session to **Normal** before you continue.
  </Step>

  <Step title="Describe the job">
    Say what should happen, which sources to use, when it should run, what needs review, and what a useful result looks like. Add this line when you want a review before creation:

    ```text theme={null}
    Ask me whatever you need. Show me the final name, prompt, schedule, and
    timezone. Wait for my confirmation before you create the Runner.
    ```
  </Step>

  <Step title="Check the final setup">
    Read the proposed name, prompt, schedule, and timezone. Starting from the main **Runners** page creates it under **No project** with no Runner-specific tool permission overrides.

    Confirm only when you're ready for the Runner to become active.
  </Step>

  <Step title="Review the saved details">
    Open the new Runner. Choose a **Project** if it needs shared background, verify **Runner model**, and review **Tool permissions** before the first scheduled run.
  </Step>
</Steps>

## Fill in the job yourself

Open **Runners**, click **New runner**, then choose **Create manually**.

Fill in:

* **Name**
* **Prompt**
* **Schedule** and **Timezone**
* **Project**, when the job belongs with other work
* **Runner model**, using **Normal** or **Ultra**

Click **Create runner** only after you've checked the schedule. The new Runner starts active.

## Set tool permissions

For connected-app writes:

* **Ask** stages an **Action Card** for review.
* **Allow** performs the action without asking.
* **Off** blocks the tool unless this Runner has its own override.

A new Runner starts with **No overrides** and uses your current **Connected Apps** settings. Add a Runner-specific override only when this one job needs a different rule. That override takes priority, including over **Off**.

Start writes on **Ask** when you're testing a new Runner. Move a specific tool to **Allow** only when you want that Runner to use it without stopping for review.

[How tool permissions work →](/connections/tool-permissions)

## Choose what carries into each run

Each run starts in a separate session. Previous run output is not automatically supplied to the next one.

Use Personal Memory for durable personal facts. Use Project context for background. Keep the job's instructions, thresholds, and review rules in the saved prompt.

The **Carry unresolved Action Cards into new runs** switch does one narrower job. It lets an unresolved **Action Card** appear in a later run. A pending **Action Card** still expires 24 hours after it was staged.

## Check the first three runs

Open the Runner and click **Run now**. The test creates a separate session and keeps you on the Runner detail page, even if the Runner is paused.

Open **Recent runs** and click the run's timestamp. Check the result, tool activity, and any **Action Cards**. Review the first three runs this way. Fix job instructions in the saved prompt and incorrect background in Project context. Then leave the schedule active or pause it while you adjust the job.

Add another Runner after the first three runs are useful and consistent.

***

## What's next?

<CardGroup cols={2}>
  <Card title="How Runners work" icon="person-running" href="/runners/overview">
    See what carries into each run and when actions wait for review.
  </Card>

  <Card title="Which AI employee should you add next?" icon="star" href="/tips/tips-and-tricks">
    Find the next job tied to revenue, cost, time, or customer delivery.
  </Card>
</CardGroup>