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

# DevTune Agents

> Run project-scoped agents, review their draft outputs, schedule repeat work, and follow run history.

**Agents** investigate your project data and produce reports, content briefs, or actions for review. They can read only their listed project tools, and every output stays a draft until a person approves or keeps it.

## Open Agents

### Refresh company context

Run **Company Context Refresher** to research your homepage, documentation, and other domains you have marked as owned in **Settings**. It compares the sources with your saved company context and proposes corrections or additions, with source links and any gaps in coverage.

Open the completed run to compare the previous context with the complete proposed text. Select **Accept and update context** to save it to **Settings › Context**, or select **Discard**. If the context has changed since the run started, run the agent again before accepting. When the research does not support an update, the agent returns a report without a replacement.

You can also start this agent from Slack. Set up scheduled runs in **Agents**. Open the linked run in DevTune to review and accept its proposal. Runs use the same credits and permissions as other predefined agents. Completed reports and proposals are available through the agent-run API and MCP tools.

### Dex chats

Dex chats in the app are private to the person who started them. Dex still reads the shared data for the selected project; other project members cannot read your chat history. Chats accessed through the API or MCP belong to the API key's creator.

To delete a chat, open the Dex conversation list, select its delete button, and confirm. This stops any work still running in that chat and removes it from your history. Billing and audit records are retained. Slack conversations continue to use their Slack channel or thread audience.

### Project agents

1. Open a project
2. Select **Agents** in the project sidebar
3. Use **Grid view** or **List view** to browse the roster

The **Agents** page has two tabs: **Agents** for the roster, schedules and runs, and **Library** for agent deliverables and uploaded source material when your plan includes it. See [Library](/library/overview).

You need the project management permission to start runs, change schedules, review outputs, or create custom agents.

## Agent roster

Each agent has one focused job and shows the project tools it can read and the draft type it delivers.

| Agent | Job |
| - | - |
| **Search Performance Analyst** | Explains weekly SEO and AEO movement and shows which search, traffic, page, and citation sources covered the report. |
| **Technical Auditor** | Explains current technical health as a baseline or compares two named audits, backed by page evidence and guard rails. |
| **Page Citability Improver** | Finds the highest-value page with citability problems and delivers a step-by-step improvement brief. |
| **404-Demand Triager** | Turns credible missing-page demand clusters into actions and states when traffic evidence is unavailable. |
| **Competitor Citation Analyst** | Finds AI answers that cite competitors or third parties for questions you should own and suggests content to compete for the citation. |
| **Brand Representation Auditor** | Checks AI descriptions against your product context and flags claims that are wrong, stale, or missing. |
| **Content Decay Detector** | Finds pages losing clicks, impressions, or citations and prioritizes refresh candidates. |
| **Question Coverage Planner** | Finds audience questions with no page to answer them and creates a brief for the largest gap. |
| **AI Crawler Access Sentinel** | Checks whether AI crawlers can reach and fetch your content and flags blocks, errors, and `llms.txt` gaps. |
| **Docs Consistency Checker** | Finds pages that contradict your product context or each other. |
| **Company Context Refresher** | Researches your owned properties and proposes updated company context for you to accept in Settings. |
| **New Page Citability Reviewer** | Reviews recently discovered pages against citability factors and briefs the fixes. |
| **Competitive Movement Digest** | Summarizes competitor visibility movement across tracked prompts and where you gained or lost ground. |

The visible roster and model choices can depend on the agent features configured for your account.

## Start a run

1. Select **Run** on an agent
2. Choose a **Model**
3. Review **Estimated per run** when history is available
4. Optionally choose a quick action or enter a focus instruction
5. Select **Start run**

DevTune checks the account credit balance before queueing. If recent completed runs exist for the same agent and model, the dialog estimates the next run from that history. The first completed run provides the first measured estimate.

## Evidence a run needs

Each agent reads a specific set of sources: probes, AI traffic sensors, Search Console, Google Analytics, your crawled content, and Audit results. The run dialog shows which of them this project has connected.

* **No connected evidence.** Nothing the agent reads is connected, so the run is stopped before it starts and no credits are charged. Connect a source named in the dialog, then start the run again.
* **Partial evidence.** Some sources are connected and some are not. The run goes ahead and its output names what it could not see.
* A source that is connected but has not delivered a measurement yet is reported separately from one that was never connected, because the two need different fixes.

A source that is connected and returned nothing is not a gap. The agent reports that as a measured zero over its window.

Scheduled runs use the same check. A blocked occurrence appears in run history with the reason and costs nothing.

The run detail shows the run log, tool steps, saved artifacts, model, trigger, timestamps, duration, and output. Status can move through **Queued**, **Provisioning**, **Running**, and **Polling** before reaching **Ready for review** or a terminal error state.

## Review the output

An agent output does not publish content or change your live site.

* Select **Add to Actions** to approve an action. It appears under **Accepted** in the [Actions queue](/insights-actions/actions-hub).
* Select **Send to Actions** to approve a content brief. Its action appears under **Accepted** with the brief attached.
* Select **Keep report** to retain a report with the run
* Select **Discard** when you do not want the draft

After a draft is ready, use **Follow-up** to ask for a revision or next step. Each follow-up starts a separate run in the same run thread, grounded in the preceding output, and consumes credits like any other run.

## Schedule repeat runs

Select **Schedule** on an agent to configure:

* **Daily**, **Weekly**, or **Monthly** cadence
* Time and, when applicable, weekday or day of month
* Model used by scheduled runs
* Estimated monthly credits when enough history exists

Scheduled runs enter the same run history and review flow as manual runs. Their outputs still wait for review.

## Run history

**Run history** combines runs across the project. Filter it by agent, then select a row to inspect its log and output. The history distinguishes manual, scheduled, and follow-up triggers and shows usage recorded for each run.

## Next steps

* [Create a custom agent](/agents/custom-agents)
* [Choose models and understand credits](/agents/models-and-credits)
* [Review credit transactions](/account-billing/credit-transactions)


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