The Agent answers with the same numbers you see in the reports and asks before changing prompts, topics or other project data; what it changes in the Brain, Memory and skills shows in the chat and can be undone. It uses the installation’s model key (Admin > Providers) and the project’s budget.
Chat#
Ask in plain language, for example “which competitor grew most in ChatGPT this month?”, and attach a PDF, a spreadsheet or an image for context. While it works, the last line of its message, after the text and artifacts so far, says what it is doing now in words, with a timer: the step under way (“Searching the web…”), or Thinking…; while it writes, a thin cursor marks the end of the text. The steps already done stay at the top, each with how long it took, and fold into “Thought for 12 s” when it ends; one that failed has a red dot and says why. A conversation still answering shows a small spinner in the list, and a dot when a new answer is waiting for you. A message sent meanwhile steers the next step. The artifact or content the chat is about stays on screen but read-only until the answer ends, under a banner that says “Editing the artifact…” with a timer and Stop, so your edits and the Agent’s never overlap; what it changed is highlighted for a moment when it lands. A short message, such as a greeting, gets a quick answer. Your conversations are listed beside the chat and only you see them. Rate an answer to teach it.
Ask Lastfind, at the bottom right of every project screen, opens the same chat in a panel beside the screen (the whole screen on a phone). It knows the page you are on, with its period, filters and the item open there, so you can ask “summarize what matters on this page” or “explain these metrics”; the answer cites the numbers that screen shows and links to it. Its conversations are the chat’s own: History lists them, and Open in Agent > Chat goes on in the full screen.
Contents and artifacts#
An artifact is what the Agent writes for you: a document, a report, a table, a brief or a content piece, each with its own icon. When you do not name a type, it makes a document, the general kind; a report is a data report, and a content piece is marketing text to publish, such as a blog post, an article for search, a landing page or a product page. Charts go inside an artifact. Artifacts with live data can update themselves, and Share creates a public link anyone can read. Contents are those pieces to publish: each has an Overview (title tag, meta description, slug and the comparison with the pages that rank today), the Content itself, an Outline and the Research it used, in a panel at the page’s right edge that folds to one button without moving the text. A content piece or an artifact opened from its list is a page of its own, without the chat: a back arrow returns to the list, the type’s icon sits before the title, and Chat opens the conversation beside it. New artifact, on the Artifacts list, starts a blank document for you to write or to ask the Agent about. In More, Width sets how wide the text reads: Narrow, Standard (an A4 sheet, the default), Wide or Full width, kept for each artifact on this device. Opportunities lists the questions where competitors appear and you do not, with Create content beside each one.
Before a content piece’s first draft, the Agent reads the Brain and searches what your own site already published on the topic, so your pages back the text as first-hand evidence. It then reads the top three Google results; Lastfind counts their words and yours, and a page it could not read is marked as not read, never estimated.
Brain and skills#
The Brain holds what the Agent must know about your brand (positioning, products, audience, voice), read on every turn; Generate with AI drafts it from your website, and you keep only what is right. Memory keeps what the team taught in conversations. Skills are repeatable ways of working, such as how to write a comparison page; the Agent can propose a new skill from work your team repeats, and you turn each one on or off. Agent > Skills lists them like artifacts, all with the same skill icon, with a search and a status filter.
What the Agent spends counts toward the project’s monthly budget, when it has one (Settings > Monitoring > Budget). The menu under your name, at the foot of the sidebar, shows what is left of the project’s budget this month, or what it spent when it has no budget.