# Agents

> Configure, schedule, run, review, and retry repeatable AI-assisted research.

Canonical URL: https://marcenta.ai/docs/agents
Markdown URL: https://marcenta.ai/site-md/docs/agents

Agents package repeatable AI-assisted work into a saved configuration. They can run once or recur, preserve run status, and produce reviewable artifacts.

## Availability [#availability]

Agents are available on Orbit unless an organization-specific entitlement enables them. The current configurable catalog contains the **Keyword Research Agent**. A catalog item without an enabled creation action should not be treated as generally available.

## Configure a Keyword Research Agent [#configure-a-keyword-research-agent]

Open **Agents**, choose **Keyword Research Agent**, and configure the research scope.

Depending on the chosen category, the form includes:

* a keyword category;
* comma-separated seed keywords or competitor names;
* optional supplemental keywords;
* a brand term where relevant;
* keywords to exclude;
* keyword intent;
* a lookback of 30, 60, or 90 days where applicable; and
* optional minimum volume and maximum difficulty filters.

Add at least one meaningful seed for non-competitor research. For competitor research, verify the company names carefully and add supplemental seeds only when they belong in the same research scope.

### Soft and strict thresholds [#soft-and-strict-thresholds]

Advanced filters default to soft thresholding. In soft mode, volume and difficulty thresholds guide the research while preserving broader coverage. Enable **Strict thresholding** only when results outside those cutoffs must be excluded.

## Choose when it runs [#choose-when-it-runs]

Priority options are displayed as **Now**, **Today**, and **Overnight**. Priority affects when the run is queued; it does not change the research instructions.

Enable **Repeat** to run the saved configuration every specified number of days. Review recurring agents whenever the product, market, competitors, or keyword strategy changes.

## Live Agents [#live-agents]

After assignment, the saved agent appears under **Live Agents** with its category, current status, priority, cadence, and latest run. Use **Run again** to start another run from the same configuration.

## Run states [#run-states]

A run can be:

* waiting to start;
* in progress;
* awaiting review;
* completed;
* failed; or
* cancelled.

Open the run to review the current stage and available outputs. Runs awaiting review require a person to inspect the result and choose the available continuation or revision action.

If a run fails, use **Retry from failed node** when available. A retry continues from the failed stage using the preserved run context; first correct invalid inputs or unavailable source data if those caused the failure.

## Review outputs [#review-outputs]

Completed or review-stage runs can create artifacts. Open the artifact from the run or from the central [Artifacts](/docs/artifacts) page.

Before using a keyword plan:

1. Confirm the category, seeds, exclusions, intents, and period.
2. Review unexpected brand or competitor terms.
3. Check low-volume and high-difficulty results against the selected threshold mode.
4. Remove irrelevant or sensitive terms.
5. Export or continue only after a human review.

## Troubleshooting [#troubleshooting]

| Symptom                         | What to check                                                                       |
| ------------------------------- | ----------------------------------------------------------------------------------- |
| You cannot create an agent      | Confirm the workspace entitlement and that the catalog action is enabled            |
| Assignment is rejected          | Add the required category, seeds or competitors, intent, and valid threshold values |
| A run remains queued            | Check its priority and allow time for earlier work to complete                      |
| A recurring run is not expected | Review whether **Repeat** is enabled and the configured interval                    |
| Output is too narrow            | Disable strict thresholding or broaden seeds, intent, and exclusions                |
| Output is irrelevant            | Tighten seeds, category, exclusions, competitors, and intent before rerunning       |

<Callout type="warn">
  Agent output is a proposal for review. Validate it before making campaign, budget, customer-facing, or contractual decisions.
</Callout>