Screenshots & figures

Articles read better with evidence in them. Each project has a library of two asset kinds the writer can embed: screenshots captured from your real product, and figures researched from real sources. It lives under Library in the app, searchable and filterable by tag, and the writer sees the whole active library every time it drafts or revises.

Demo credentials

Screenshots need a way in. Under Library settings you store a login URL and demo credentials for your product; the password is encrypted at rest (AES-256-GCM) and only the capture worker can decrypt it. Use a dedicated demo account with realistic-looking data — every capture sees exactly what that account sees.

Discovery

Saving credentials the first time triggers discovery: the worker signs in and walks your app's navigation, and Contentcron picks the pages worth showing — up to about a dozen — then names and tags each one by feature. They land in the library as active entries and are captured right away. A Rebuild button in settings re-runs discovery whenever your product has changed shape.

Staying current

• Every active screenshot is re-captured daily. The new capture is pixel-diffed against the stored raw, and the published image only updates when the UI actually changed — styling noise doesn't churn your articles.
• Each screenshot lives at a stable public URL. Articles embed by URL, so when a shot updates, every article that uses it — published or draft — shows the new UI at once. Ship a redesign and your archive follows.
• Deleting a screenshot that a published article embeds warns you first and lists the affected articles.

The writer requests what it needs

The writer embeds fitting shots from the library on its own. When an article calls for a page that isn't in the library yet, it files a screenshot request mid-draft; the worker captures the route, and a follow-up revision embeds it once it exists. Requests are deduped against the library and against the article's own earlier requests, and you can also ask for one directly in the article chat.

Figures

Figures are researched with live web search, and one rule is enforced structurally: a chart or stat that cannot cite a source is rejected. The research pass is allowed to return nothing rather than invent a number.

Image figures embed with an italic, linked source caption; stat callouts embed verbatim. Any figure can be regenerated from its detail view if the data or your palette moved on.

Your palette, not a template's

Figure colors are extracted from your site once and cached; every figure renders in that palette so charts look like yours. You can override the colors in Library settings — the override applies to figures created after the change, since each figure snapshots its palette at creation.

Starter figures

An empty library has a cold-start problem, so Generate starter figures proposes up to eight figure ideas for your niche in one pass. They arrive as suggestions: approve one and the research runs, renders, and files it into the library; reject the rest and they disappear. Approvals run in the background and in parallel — you can approve several and keep working.

How assets reach articles

• The writer and reviser see every active screenshot and figure, with names, routes, and tags, and embed by exact URL.
• Mid-draft, the writer can request up to two new figures per article; like screenshots, they're researched and embedded in a follow-up revision.
• The article chat can embed an existing asset or request a new one on instruction — see the refinement chat.