Research, images & documents
Beyond your CMS and code, every workspace includes a set of built-in skills. They run on platform-provided services — nothing for you to sign up for — and their usage is metered in credits like everything else.
Included skills
| Skill | What it does |
|---|---|
| Web research | Search the live web and read pages — for facts, sources and current information that isn't in your CMS. |
| Stock images | Search free professional photography and import a chosen photo straight into your CMS asset library, credit included. |
| Image generation | Generate original images from a description, ready to place in content or save to the library. |
| PDF documents | Produce finished PDFs — one-pagers, reports, print pieces — stored in your Content library. |
Web research
The assistant searches, reads the pages behind the results, and works from what they actually say — sources are linked so you can check them.
The same skill powers page screenshots: the assistant can capture any live page — including your own site on staging after a change, desktop or mobile — see the picture itself, and share it in the chat. That's how it visually double-checks design work before showing you the result.
Stock images & generation
When content needs a picture, the assistant first looks for a suitable existing asset in your library. Failing that it can search stock photography — you pick from candidates shown inline in the chat, and the chosen photo is imported into your CMS with the photographer credit kept — or generate an original image to your brief. Imports go into the same environment as the document that uses them.
PDF documents
Ask for a document — a case study one-pager, a proposal, a print-ready piece — and the assistant designs and renders it as a PDF. Finished files land in your Content library and open in the chat's preview pane; you can download or share them from there.
Figma design integration
On Growth and above, connect your Figma account and the assistant can work straight from your designs: paste a Figma link in chat — ideally to a specific frame — and it reads the frame's exact spacing, colours, typography and layout values, sees a visual render of it, and builds the implementation from those values rather than a description. If the link points at a whole file, it lists the pages and frames so you can pick one.
- Connecting: a team admin adds a Figma personal access token (Figma → Settings → Security → Personal access tokens) under Integrations in the portal — or during onboarding. The token is verified live and stored encrypted; it can be updated or revoked from the same page at any time.
- Read-only: the assistant can never modify, comment on, or delete anything in Figma. Scope the token to File content: Read-only.
- Access scope: a Figma token grants read access to every file its owner can open — Figma tokens cannot be limited to a single project. If your Figma team holds work for other clients, connect a dedicated Figma account that only has access to your own projects.
- No Figma? You can always attach an exported HTML file or a screenshot of a design directly in chat instead — see working from design files.
Getting the best results
- Link to the frame, not the file. In Figma, select the frame you want built, then copy the link (right-click → Copy link to selection). That puts a
node-idin the URL pointing at exactly the right thing. A link to a whole file or page still works — the assistant lists the frames and asks which one you mean — but a targeted link skips that round trip. - One section per request. A full page design is a lot of detail; asking for it in one go means the assistant works from a compressed view. You'll get noticeably more faithful results building section by section — "build the hero from this link", then the next frame — just as a developer would.
- Name your frames and layers. The assistant reads layer names as part of the design. "Hero", "Pricing card — featured" and "CTA button" guide it far better than "Frame 412" and "Rectangle 49".
- Structure helps. Designs using auto-layout, text styles and colour styles come across with clean, exact spacing and typography values. Absolutely-positioned mockups still work — the assistant gets coordinates and sizes — but well-structured files translate more precisely.
- Say where it goes. Pair the link with the destination: "build this as a new section on the homepage above the footer" or "restyle the pricing cards to match this frame". The design says what it looks like; you say where it lives.
- Prototype links work too. Links copied from presentation/prototype mode (
/proto/…) are read the same way as design links. - Iterate in chat. After the first build, ask for adjustments conversationally ("tighten the spacing between the cards", "the heading should be the larger size from the design") — the assistant still has the design's values in the conversation.
Bring your own keys
Enterprise workspaces can supply their own API keys for any of these services (for compliance or billing reasons) via the admin console — a workspace's own key always takes precedence over the platform's. For everyone else, the built-in services just work.