An SEO agent that shows its work
Ask in plain language. The agent plans the steps, calls what it needs from nineteen tools, and works through up to eight of them before it answers. Its reasoning streams into the transcript while it runs, so you can see which tool it reached for and what came back.
The site record names the domain, so that lookup comes first. Then the audit. Core Web Vitals only if performance is what scored badly.
What the agent has to work with
Nineteen tools, up to eight steps a turn
One question can turn into a page audit, a Core Web Vitals fetch, a screenshot and a read of your own site record before a single word comes back. The agent picks the order. It keeps working after a tool returns, so it can read a result and then act on it in the same turn.
Reasoning you can read while it runs
Extended thinking is switched on with a 4,096-token budget and streamed into the chat as it happens, in a panel you can open. When the agent decides an audit is the wrong tool for your question, you see it decide.
It reads your account
Your registered sites, the crawl and issue history on each one, your saved documents with their SEO metrics, and your recent activity across the platform. Five of the nineteen tools do nothing but fetch rows that belong to you, so answers about your work come from your work.
Tools that name what they did not measure
The keyword gap tool returns a field stating whether search rank and volume were measured at all, and its own description forbids the model from calling the result a ranking until they were. The backlink tool names the provider credential a deployment has to hold before any backlink figure exists to report. A tool that cannot answer says which piece is missing.
How it works
- 1
Ask about a site you own
Type the question the way you would say it out loud. The agent can look your sites up itself, so getting started does not require pasting a URL or knowing which tool the question maps to.
- 2
Watch it choose
Reasoning arrives first, then the tool calls, each labeled with what it ran and what came back. Screenshots render in the transcript. Accessibility findings come back drawn on top of the shot they were found in. Core Web Vitals arrive as a labeled chart.
- 3
Take the output, keep the thread
Meta titles and descriptions, JSON-LD for nine schema types, Next.js CMS integration code and drafted posts come back as text you can carry into the surface that uses it. Conversations are saved, so you can reopen one next week and pick up where it stopped.
Put an SEO agent on the work
Seven-day trial. Ask it something about a site you actually own and read the reasoning before you take the answer.