Skip to main content
Browse documentation

Product

AI Agent

The agent answers questions about your brand using your data, by calling the same read paths the dashboards use. It is not a general chatbot with your logo on it — every claim it makes traces back to a tool call you can see.

Where to find it

  • The AI Agent page — full width, with a rail of past conversations you can rename or delete.
  • In place, anywhere in the app — press ⌘J (Ctrl-J on Windows and Linux) to open it as a panel without losing the view you are on.

What it can read

The read tools are exactly the surfaces you already have, reachable in conversation instead of by navigation:

ToolWhat it fetches
get_visibility_rankingThe visibility leaderboard — you against your competitor set.
get_visibility_timeseriesVisibility over time.
get_change_eventsWhat changed, and when.
get_mentionsIndividual brand mentions in model answers.
get_sourcesThe URLs the models cited.
get_sentimentHow you are being talked about.
get_knowledge_baseYour own brand profile, tone, entities and documents.
get_site_healthSite health and AEO score.
get_onpage_issuesThe on-page issue backlog.
get_prompt_performanceHow individual prompts are performing.
get_traffic_summarySearch traffic from your connected providers.
fetch_web_pageA page on the open web, when you ask it to look at one.

The tool ledger

As the agent works, each call appears as a plain-English line — Pulled the visibility leaderboard, Read sentiment — with the raw tool name beside it in monospace. You can see what it looked at before you decide whether to believe the answer.

Actions are proposed, never silently performed

Six tools write rather than read: draft_content, create_llms_txt, fix_issue, activate_prompts, add_competitor and publish_content. These never appear as a completed ledger line. They render as a confirmation card showing precisely what will happen, and nothing is written until you accept it.

If the server has no AI model configured, the agent page says so in a banner at the top rather than failing question by question. That is an administrator setting, not something a workspace can fix from the UI.