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

# Background agents

> Set up agents that run on a schedule and surface findings for you to review

Background agents are Catalyst runs that execute on a schedule without you sending each message manually. An agent works over your team's tickets, workflows, databases, and connected apps on a recurring cadence, then surfaces what it found as a **finding** you can review, act on, or dismiss.

Use background agents for recurring, proactive work: summarizing the last day of tickets, scanning your device fleet for at-risk endpoints, reclaiming unused licenses, or spotting automation opportunities across your ticket volume.

<Frame caption="The Agents view listing scheduled runs with their instructions, schedules, and status">
  <img src="https://mintcdn.com/serval/P5n-CldG9r_xGwlj/images/serval-docs/catalyst/catalyst-agents-list.png?fit=max&auto=format&n=P5n-CldG9r_xGwlj&q=85&s=53dee2b284979711ae24c068fc774ca1" alt="Catalyst Agents view showing scheduled agents with instructions, schedules, and status" width="1870" height="772" data-path="images/serval-docs/catalyst/catalyst-agents-list.png" />
</Frame>

## How background agents work

1. The agent runs on its schedule (or when you run it manually).
2. Each run is a Catalyst session you can open to see exactly what the agent did.
3. When a run produces something worth reviewing, it becomes a **finding** in the Agents feed.
4. You review the finding and apply, propose, dismiss, or acknowledge it.

<Note>
  A run only becomes a finding when it produces reviewable substance: an action the agent took or a highlight worth surfacing. Runs that find nothing actionable complete without creating a finding.
</Note>

## Create an agent

You can start from a blank agent or install a prebuilt one.

<Steps>
  <Step title="Open the Agents view">
    Go to **Catalyst → Agents**.
  </Step>

  <Step title="Start a new agent or install one">
    Choose **New agent** to write your own, or pick a prebuilt agent from the marketplace (see [Installable agents](#installable-agents)). Installed agents come with a ready-made prompt and schedule you can edit.
  </Step>

  <Step title="Configure the agent">
    Fill in the fields below, then save.
  </Step>
</Steps>

### Agent settings

<ParamField path="Name" required>
  How you identify the agent in the list, for example "Daily ticket review."
</ParamField>

<ParamField path="Prompt" required>
  What the agent should do each run. Be specific about what to look at and what to produce, for example: "Investigate open tickets and summarize anything concerning."
</ParamField>

<ParamField path="Schedule">
  How often the agent runs: **Hourly**, **Daily**, **Weekdays**, **Weekly**, or **None** for a manual-only agent you run on demand. Daily, weekday, and weekly schedules let you set the time of day. Weekly also lets you pick the day. Start times may vary by a few minutes.
</ParamField>

<ParamField path="Run as">
  The user whose permissions the agent uses while it runs. The agent can only see and do what that user can, so choose someone with the access the agent needs.
</ParamField>

<ParamField path="On completion">
  An optional action to take when the run finishes: **Run a workflow**, **Create a ticket**, or **Auto-publish changes** (publish the run's staged changes without a manual review step).
</ParamField>

<ParamField path="Notifications">
  Optionally notify a user when the agent produces a report (**Report ready**), needs input to proceed (**Action required**), or finishes regardless of outcome (**Session completed**).
</ParamField>

<Warning>
  The **Run as** user determines the agent's permissions on every run. If that user's access changes, the agent's capabilities change with it.
</Warning>

## Installable agents

Instead of writing a prompt from scratch, install a prebuilt agent for a common use case and customize it. Each ships with a recommended schedule you can change.

| Agent                                       | Cadence | What it does                                                                                                                                    |
| ------------------------------------------- | ------- | ----------------------------------------------------------------------------------------------------------------------------------------------- |
| **Daily Ticket Digest**                     | Daily   | Summarizes new and unresolved tickets from the last 24 hours and highlights what needs attention, with links back to each ticket.               |
| **Service Desk QA / Quality Review**        | Daily   | Grades how human agents handled escalated tickets on clarity, correctness, comprehension, and collaboration, and produces a coaching scorecard. |
| **Proactive Device & Endpoint Remediation** | Daily   | Scans the device fleet for at-risk endpoints, flags emerging issues, and drafts remediations for you to review.                                 |
| **Ticket Trend & Volume Analytics**         | Weekly  | Tracks week-over-week ticket volume by category and surfaces the most common and critical failure patterns.                                     |
| **Automation Opportunity Finder**           | Weekly  | Reviews recent tickets for repetitive, manual requests with no coverage and suggests workflows, guidance, or knowledge articles.                |
| **Proactive License Reclamation**           | Weekly  | Finds unused application licenses, reports the savings, and drives reclamation through your deprovision workflows.                              |

<Tip>
  When an installed agent gets a new version, its entry shows **New version available** so you can pull in the updated prompt.
</Tip>

## Review findings

Each finding is the agent's pitch for a single run. Open one from the Agents feed to see:

* **Highlights**: the key points on why the run matters.
* **Activity**: the concrete actions the agent took during the run.
* **Sources**: the tickets, resources, and web pages the agent drew on.
* **Changes to review**: any staged changes (workflows, skills, knowledge) the agent drafted, ready to inspect before publishing.

From a finding you can:

* **Apply** the suggested changes, or **Propose for review** on teams that require approval before publishing (see [Changes and reviews](/sections/documentation/catalyst/changes-and-reviews)).
* **Dismiss** the finding when there are staged changes you don't want.
* **Acknowledge** the finding to clear it when there is nothing to apply.
* **Chat about this** to open the agent's session and continue the work in Catalyst.

Findings you have already applied, dismissed, or acknowledged move to **Past findings**, where you can revisit them and open the underlying run.

## Manage agents

From the Agents view you can **pause**, **resume**, **edit**, and **delete** any agent. Pausing stops the schedule without deleting the agent or its history. Every run, scheduled or manual, creates a Catalyst session, so you can always open a run to inspect exactly what the agent did.

## Learn more

<CardGroup cols={2}>
  <Card title="Catalyst workspace" icon="layout-panel-left" href="/sections/documentation/catalyst/catalyst-workspace">
    The Chat, Agents, and Catalyst skills views
  </Card>

  <Card title="Changes and reviews" icon="git-pull-request" href="/sections/documentation/catalyst/changes-and-reviews">
    How staged changes are reviewed and published
  </Card>

  <Card title="Building workflows" icon="wrench" href="/sections/documentation/catalyst/building-workflows">
    Build automations with Catalyst
  </Card>

  <Card title="Permissions" icon="shield-check" href="/sections/documentation/platform/permissions">
    Who can create and manage agents
  </Card>
</CardGroup>