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

# Build your own agent

> Start from a ready-made agent or from scratch, then customize the runtime, credentials, skills, repo, project, and schedule — and save it as a preset.

<Info>
  Managed agents are available on **Enterprise**. Through **September 15, 2026**, Signal and managed agents are included at no additional charge for Enterprise. See [Pricing and usage](/docs/ax/security-and-settings/pricing-and-usage).
</Info>

Every managed agent starts the same way: **Signal & Agents → + New Agent**. From there you either pick a ready-made agent and adjust it, or start from scratch and write the task yourself.

<Frame>
  <video
    src="https://storage.googleapis.com/arize-assets/doc-images/managed%20agents/managed-agents-agent-studio.mp4"
    alt="Starting a managed agent from the New Agent gallery"
    width="100%"
    height="100%"
    style={{
  display: 'block',
  objectFit: 'fill',
  backgroundColor: 'transparent',
}}
    controls
    autoPlay
    muted
    loop
  />
</Frame>

## Start from a ready-made agent

The gallery groups agents by what they are for. Picking one pre-fills its task prompt and tells you which controls still need setting.

| Group                  | Agents                                                       |
| ---------------------- | ------------------------------------------------------------ |
| **Debug & fix**        | Investigate failing traces · Fix a bug in your repo          |
| **Incident & on-call** | SRE / On-Call · Incident Commander · Triage a monitor alert  |
| **Monitor & optimize** | [Cost Agent](/docs/ax/agents/cost-agent) · Recurring health check |

Every setting a ready-made agent pre-fills remains editable before launch.

<Note>
  A **Security agent** covering PII, secrets, and prompt-injection review is built but not yet in the gallery, along with **Prompt Optimization** and **Eval / Quality** agents. They appear once their access roles land.
</Note>

## Or start from scratch

**Start from Scratch** opens **Agent Studio** — the composer — with nothing pre-filled. You choose the harness, sandbox, project, skills, and task prompt yourself, then **Run session** or **Setup as automation**. Use it when none of the ready-made agents match the job.

Every run is configured in Agent Studio. A ready-made agent fills it in through the guided setup below; Start from Scratch opens it empty.

## Guided setup

Choosing a ready-made agent walks you through up to three steps instead of dropping you straight into Agent Studio. Steps you do not need are skipped — an agent that reads only Arize traces has nothing to connect, so it drops the first step entirely.

<Steps>
  <Step title="Connect">
    Connect the outside systems this agent needs — **GitHub** to clone a repo and open PRs, **Datadog** to pull metrics and logs. When a GitHub skill is connected, you set the repository here too, since a repo does nothing without it. Arize trace access is provisioned automatically and never appears here.
  </Step>

  <Step title="Configure">
    Bind the **tracing project** the agent reads, and choose whether it runs as a one-off **session** or a recurring **automation**. For an automation, name it and set the trigger.
  </Step>

  <Step title="Review & launch">
    Edit the task prompt, confirm the setup, and start it.
  </Step>
</Steps>

On the first run you may need to acknowledge that Claude Code can process project traces, your instructions, and connected repo or tool data.

## Session or automation

| Mode           | Use for                                                                                     |
| -------------- | ------------------------------------------------------------------------------------------- |
| **Session**    | One-off work. Follow the transcript and send follow-ups to steer it.                        |
| **Automation** | The same configuration on a **cron** schedule or a **metric** threshold. Requires a preset. |

Some agents support only one mode. *Fix a bug in your repo* is session-only, since it acts on a bug you describe. *Triage a monitor alert* and *Recurring health check* are automation-only, since each is defined by its trigger.

[Signal](/docs/ax/observe/signal) uses the same automation model, enabled per tracing project.

## What you can customize

| Setting               | Options                                                                                                                                                 |
| --------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Harness**           | The agent runtime inside the sandbox. **Claude Code** today; others on the roadmap.                                                                     |
| **Sandbox**           | The compute environment. **Arize**-managed Kubernetes today; Daytona and Vercel on the roadmap.                                                         |
| **Agent integration** | Which Anthropic credential powers the harness — Arize-managed by default, or your own key. See [Connect your harness](/docs/ax/agents/connect-your-harness). |
| **Skills**            | GitHub, Datadog, broader Arize API access, or a custom skill from your own repo.                                                                        |
| **Repo**              | The GitHub repository to clone. Requires a GitHub skill.                                                                                                |
| **Project**           | The tracing project whose traces and evals the agent may read.                                                                                          |
| **Task prompt**       | What you want done. Pre-filled by ready-made agents, fully editable.                                                                                    |
| **Trigger**           | Cron schedule or metric threshold, for automations.                                                                                                     |

## Save it as a preset

A **preset** is a saved bundle of harness, sandbox, optional agent integration, optional skills, and optional repo — so you configure it once instead of per run.

Manage presets under **More → Agent Presets**, or create one inline while setting up an agent.

<Frame>
  <video
    src="https://storage.googleapis.com/arize-assets/doc-images/managed%20agents/managed-agents-create-preset.mp4"
    alt="New Agent Preset form showing harness and sandbox selection with optional skills and GitHub repo fields"
    width="100%"
    height="100%"
    style={{
  display: 'block',
  objectFit: 'fill',
  backgroundColor: 'transparent',
}}
    controls
    autoPlay
    muted
    loop
  />
</Frame>

Two things to know about presets:

* **Automations require one.** A one-off session can configure everything ad hoc; a recurring automation inherits its whole setup from a preset.
* **Some are managed for you.** Attaching a repo in **Configure Signal** creates a Signal-managed preset behind the scenes. Those stay out of the general preset list.

You pick the project each time you launch, not on the preset, so one preset serves many projects and setting it up for a different project does not duplicate it.

## Track what is running

Sessions and automations both land in **Signal & Agents**:

* The **Managed Agents** tab shows recent runs grouped by preset and colored by status. Click a run to open it.
* The **Agent List** tab is a searchable table of every session and automation, with its preset, prompt, status, and type.
* The **Automations** tab lists recurring automations, where you can enable, disable, or delete each one.

From a run you can read the live transcript, send follow-ups, jump to a PR the agent opened, view the agent's own traces, reprovision a finished session to continue it, or archive it. Multiple sessions run in parallel.

Opening an automation shows its trigger, prompt, preset, recent runs, and any PRs it produced. Pause, resume, or delete it from there.

## Outputs

| Output                    | When                               |
| ------------------------- | ---------------------------------- |
| Investigation markdown    | Always possible                    |
| Eval labels, dataset rows | Task-dependent                     |
| Branch or PR              | When the run has repo write access |

Code changes arrive as a pull request for you to review and merge. A managed agent does not deploy to your production environment, and is not a runtime for your customer-facing app.

## Related docs

* [Cost agent](/docs/ax/agents/cost-agent) — The ready-made agent for spend
* [Signal](/docs/ax/observe/signal) — Ranked issues on a tracing project
* [Connect your harness](/docs/ax/agents/connect-your-harness) — Credentials, skills, permissions, and boundaries
* [Arize Skills](/docs/ax/skills/overview) — A different thing: skills in your own local coding agent