Skip to content
QuantCDN

AI assistant

Studio’s AI assistant understands your content schemas, the current entry you’re editing, and the rest of your project. It can create and edit content, modify source files, update styling, change site settings, and manage pull requests — all through natural conversation backed by a tool registry that the model can call directly.

Opening the assistant

Section titled “Opening the assistant”

The AI assistant is available across Studio — from the collection browser, the visual editor, the code editor, and the pull-request review screen. Open it from the sparkle icon in the Studio nav rail (tooltip AI with ⌘J), or press ⌘J at any time.

The assistant automatically receives context for the surface you’re on:

  • Current branch — so any changes commit to the right branch
  • Active collection schema — your Zod schema, serialised as JSON Schema, for the collection you’re viewing
  • Active entry — when you’re editing an entry, the frontmatter and Markdown body are passed in
  • Pull-request context — when you’re on a PR review screen, the PR number is in scope so the assistant can read comments, list changed files, and post replies
  • Edge function context — when you’re editing an edge function, that scope is included

You can also drop or paste images directly into the chat. The model uses them to inform layout, copy, or component generation — for example, “make this hero look like the attached screenshot”.

What the assistant can do

Section titled “What the assistant can do”

The assistant has access to 27 tools across eight categories. Each turn may chain multiple tool calls — up to 25 rounds of tool execution per user message — so a prompt like “create a blog post about edge functions and open a pull request” can trigger several tools in sequence before the assistant replies.

Content tools

Section titled “Content tools”
Tool Description
create_entry Create a new entry in a collection with frontmatter and body. Pass a locale to write into a locale subdirectory.
update_entry Merge frontmatter updates into an existing entry, optionally replacing the body.
list_entries List slugs in a collection.
get_entry Read the full frontmatter and body of an entry.
patch_entry_body Find-and-replace within an entry’s Markdown body — preferred for targeted edits.
delete_entry Remove an entry from a collection.
insert_island Embed an edge function in an entry as a <quant-island> element with a loading directive (load, visible, idle, or media).

Schema tools

Section titled “Schema tools”
Tool Description
list_collections List collections with entry counts, paths, fields, and Studio compatibility.
get_schema Return the JSON Schema for one collection (or all).
create_collection Add a new collection — updates src/content/config.ts, creates the data directory, and scaffolds a detail page template.

File tools

Section titled “File tools”
Tool Description
read_file Read any file in the project (truncated at 10,000 characters).
write_file Create or overwrite a file. Rejects TypeScript inside <script> blocks in .astro files.
patch_file Find-and-replace in any file.
list_files List files matching a glob pattern.
create_directory Add an empty directory (placed via a .gitkeep).

Media tools

Section titled “Media tools”
Tool Description
list_media List media assets, optionally scoped to a subdirectory.
upload_media Upload a media file from base64-encoded content into public/.

Style tools

Section titled “Style tools”
Tool Description
update_tailwind_config Patch or replace tailwind.config.mjs / .cjs.
update_global_styles Patch or replace the project’s global stylesheet (default: src/assets/styles/global.css).

Workflow tools

Section titled “Workflow tools”
Tool Description
create_pull_request Open a PR from the current branch into main.
list_pull_requests List PRs filtered by state (open, closed, all).

Review tools

Section titled “Review tools”
Tool Description
get_pull_request Read PR details — title, body, state, head, base.
get_pr_comments Read all comments on a PR.
add_pr_comment Post a comment on a PR.
get_pr_files List files changed in a PR with addition/deletion counts.

Settings tools

Section titled “Settings tools”
Tool Description
get_site_settings Read the project’s i18n configuration and brand identity (colours, fonts, tone of voice, logo).
update_site_settings Update the i18n section (locales, default, URL prefix) or the brand section (colours, fonts, tone of voice, logo). The i18n update also rewrites astro.config.*.

Example prompts

Section titled “Example prompts”
  • “Write a blog post about edge computing” — creates a new entry via create_entry using defaults from the blog schema.
  • “Update the hero image on my latest post to /images/new-hero.jpg” — uses update_entry to merge the new value into the entry’s frontmatter without touching anything else.
  • “Add a category field to the blog schema with options tutorial, news, opinion” — uses create_collection or schema-update tooling to extend the Zod definition.
  • “Change the primary colour to #007c64” — uses update_tailwind_config or update_global_styles depending on where colour tokens live.
  • “Enable French and Spanish translations, with English as the default” — uses update_site_settings to update astro.config.* and the project i18n config.
  • “Open a pull request titled ‘Translation updates’ with these changes” — uses create_pull_request to raise a PR from the working branch into main.
  • “Read the open review comments on PR 42 and address them” — chains get_pr_comments, edits via the appropriate content/file tools, then add_pr_comment.

How it works

Section titled “How it works”

Each user message starts a tool loop. The model decides which tools to call, Studio executes them against your project, and the results are fed back into the conversation. The loop stops as soon as the model returns plain text (no tool calls) or after 25 rounds, whichever comes first.

When the assistant mutates content, styles, structure, or settings, Studio dispatches a refresh event so the active editor or preview reflects the change without a page reload.

You can choose which model the assistant runs on from a dropdown in the assistant header. The list is filtered to the tool-capable models your organisation has approved via AI governance; if no list is configured, the org’s default tool model is used.

Next steps

Section titled “Next steps”