An AI teammate that knows your content

Root AI is a chat assistant in the CMS. Ask it a question (“Which pages still mention the winter sale?”) or give it a job (“Translate the spring harvest page into German and French and add it to the spring release.”). It works through the same actions a person would, and shows each step it takes.

Every chat runs in one of three modes, chosen by the person using it:

• Read only: Root AI can look at content, but cannot change anything.

• Ask before writing: it plans changes and waits for approval before editing a draft. This is the default.

• Auto-apply draft edits: it edits drafts on its own.

In every mode, Root AI works within the permissions of the person using it, only changes drafts, and never publishes. Publishing always goes through your normal workflow and checks.

AI also appears where editors already work: rewriting a headline, describing an image for accessibility, summarizing what changed before a publish, or translating missing strings. Each suggestion is reviewed before it is saved.

Root AI is not tied to one vendor. Teams connect the models they have approved, from providers such as Anthropic, OpenAI, and Google, or a self-hosted model, using their own accounts and keys. Editors pick a model in the chat, and admins decide which ones are offered.

For teams using AI agents in their engineering workflow, Root.js includes a command-line tool and agent skills that let an agent propose content changes. Proposals are reviewed and tested like code changes before they are applied, so automation never bypasses review.