Working with the assistant / Chat basics

Chat basics

Your team works with the assistant at app.agenticwebteam.com — a chat, plus everything you need to see what it's doing and what it produced.

The portal

  • Conversations are saved. Recent chats live in the sidebar; open any of them to continue where you left off. Start fresh with New chat.
  • Ask in plain language. No commands or syntax — describe the outcome you want. If a request is ambiguous, the assistant asks one clarifying question rather than guessing.
  • Long jobs are fine. The assistant runs in the background — it can write a full article or restructure a section in one turn, and the conversation streams its progress.
  • Shortcuts in the sidebar: your Content library (files the assistant produced), Approvals, Team & Roles and Usage & Billing (admins), and the master Console (platform admins).

Watching it work

Every tool action shows as a step in the reply — "Reading content…", "Updating document…", "Committing change…" — ticking off as it completes. Steps that touched a document are clickable: view › opens the document in the preview pane so you can see exactly what was changed.

Working in parallel

You don't have to wait for a long job. While a task runs, start a New chat and set another one going — tasks run at the same time. Chats with a running task show a pulsing dot in the sidebar; when a background task finishes, a ✓ appears on the chat and a toast pops up with a View button. You can also keep typing in a busy chat: the message queues (shown above the composer, removable) and sends itself the moment the current task completes.

Two parallel tasks can safely work on different parts of the site. Avoid pointing them at the same page or document at the same time — the later write wins.

Attachments

Attach images, PDFs, text files or zip archives (up to 4 MB each, 10 per message) with the 📎 button, by pasting, or by dragging into the composer. The assistant reads them — brand guidelines, a spreadsheet export, a screenshot of a bug, a draft document — and works from their contents. For a zip, it sees the file list and reads just the entries it needs, including images inside the archive.

Working from design files

Attachments are also how you hand the assistant a design to implement. Export the design from your design tool — a zip export from Claude Design (attach it whole, no need to unpack), a page saved from Figma's dev mode, or any HTML/CSS handoff file — attach it, and ask for it to be built. The assistant reads the full markup, styles and packaged assets and implements from them, so spacing, colours and type come from the file rather than a description. A screenshot or PDF of the design works too: images give the assistant the visual target, while HTML exports give it the exact values — attaching both gets the best results.

On Growth and above there's a faster route for Figma: connect your Figma account once (portal → Integrations) and then just paste a Figma link in chat — no export needed. See Figma design integration.

Attached files are re-read on every turn of that conversation, so a large design file increases the credit cost of a long session. Attach it, get the build done, and start a fresh conversation for unrelated work — or, if a design should be a standing reference for the whole team, commit it to your connected repository instead and the assistant will read it on demand.

The preview pane

Files and documents open beside the chat, not in downloads: images and generated pictures render directly, PDFs get an embedded viewer, markdown and text render readably, and CMS documents show as a field-by-field view. Click any file link in a reply, an attachment chip, or a document step to open it; Esc closes. The pane opens at half the window and you can drag its left edge to resize — your width is remembered.

Previewing your site

The pane also shows your live site. Ask the assistant to "preview the page" and it opens the page beside the chat itself; every page link in a reply also carries a Preview ⧉ chip you can click to reopen it. After a visual change, the assistant offers a preview as standard — typically the staging version, so you approve before anything is promoted to production.

  • Screenshots in the chat — the assistant can capture a page (desktop or mobile) and show the picture in the conversation, checking its own work visually before showing you. (Growth and above — uses the web research skill.)
  • Mock previews — want to see a design before anything changes? The assistant builds the proposed page as a standalone mock that opens in the pane, with nothing committed until you approve.
  • If a site blocks embedding — some sites' security headers stop them displaying inside another page. The pane offers open-in-new-tab instead, and the assistant can always show a screenshot or mock in its place.

The environment switch

The Production/Staging control (top bar and composer) sets the suggested default target for new requests. The assistant still confirms the target and draft-vs-published choice before the first write of any task — and your stated preference then covers the rest of that task. See Environments.

The context rail on the right shows what the workspace is connected to, this session's stats, and recent write activity across the team.