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

# Agent setup

> What each setting on the Agent Setup tab is used for, and why to keep the system prompt and workflow current.

<Info>
  Agent Setup is the first tab of **Agent Settings** for the selected agent. You fill most of it in when you [create the agent](/docs/getting-started/quickstart).
</Info>

Agent Setup tells Tuner what your agent is and what it's supposed to do. Tuner uses these settings to score calls, generate simulation Scenarios and diagnose failures, so they matter well beyond the agent's name.

## What each setting is used for

| Setting | What Tuner uses it for |
| - | - |
| **Agent name** | How the agent is shown across Tuner, for example in the agent selector. |
| **Primary Goal of this Agent** | A one-sentence summary of what the agent is for. The [Tuner MCP server](/docs/mcp/setup) reads it, and writes it when you set up the agent through MCP. |
| **Provider** and **Agent Remote ID** | How calls are matched to this agent. Calls sent from the SDKs or the API carry the Agent Remote ID. See [API keys and IDs](/docs/getting-started/api-keys#find-your-agent-remote-id). You can copy the ID here but not change it. |
| **Agent Language** | The default language and accent of the Simulated Caller when you [run a simulation](/docs/simulation/run-a-simulation). You can still change it for each run. |
| **Call Direction** | Whether Tuner tests the agent with [inbound](/docs/simulation/inbound/setup) or [outbound](/docs/simulation/outbound) simulation. The **Simulation Setup** tab changes to match. |
| **System Prompt / Instructions** | The reference for what your agent is supposed to do. Used by evals, simulation and diagnosis (below). |
| **Workflow (JSON)** | Optional. The conversation flow as steps, each with its own instructions. Used the same way as the prompt. Leave it empty if your agent doesn't use a workflow. |

## Why the prompt and workflow matter

Tuner judges and tests your agent against the prompt and workflow saved here, not the ones running on your platform.

* **Evals:** a [custom eval](/docs/evaluation/custom-evals) can read the system prompt and workflow when you add them under **Inputs Used**, so the judge scores the call against your actual instructions.
* **Simulation:** generated Scenarios are written from your prompt and workflow, so the Simulated Callers test what your agent is meant to handle.
* **Diagnosis:** [MCP diagnosis](/docs/mcp/diagnose) compares failing calls against them to find the root cause.

If they're out of date, evals score against old rules, simulations test old behavior, and diagnosis blames the wrong instructions. If they're empty, all three have nothing to compare against.

## Keep them current

Update this tab every time you change your agent's prompt or workflow in production.

**Retell and Vapi agents:** when you pick a Retell or Vapi agent in the wizard, Tuner copies its system prompt from the provider once. It doesn't pull later changes, and saving here doesn't change the prompt on Retell or Vapi. When you edit the prompt on your provider, paste the new version here too.

To edit, open **Agent Settings > Agent Setup**, change the fields and click **Save Changes**. To update these settings from code, use [Update Agent](/api-reference/agents/update-agent).

## Next steps

<CardGroup cols={2}>
  <Card title="Custom evals" icon="list-check" href="/docs/evaluation/custom-evals">
    Write evals that score calls against your system prompt and workflow.
  </Card>

  <Card title="Diagnose your agent" icon="stethoscope" href="/docs/mcp/diagnose">
    Find why calls fail by comparing them against your prompt.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.