Manual-first by design
Jakka does not auto-generate recommendations every time you scan. With one exception (the first homepage scan on a new project), every generation is a deliberate click. This protects your token balance.
Where the Generate button lives
AI Studio > Recommendations: the main entry point. Pick scope (Priority Pages, Specific pages, or All).
Project Overview: the AI Recommendations widget shows context readiness and a Generate Recommendations button.
Test Run drawer: generate recommendations grounded in a specific run.
Page detail drawer: generate recommendations for one page.
Context indicators
Before you generate, the page shows four indicators:
Brand Kit: Completed or Incomplete.
Project Brief: Completed or Incomplete.
Discovered pages: how many Jakka has crawled.
Page Content: Available means the AI has full page text to work with.
Generation works even when context is incomplete, but accuracy drops. Aim for Completed on Brand Kit and Project Brief before generating at scale.
Speed selector
Pick Fast or Slow. Fast uses a smaller model pass and lower token cost. Slow uses deeper reasoning, costs more, and tends to produce sharper, more specific recommendations. The token estimate updates with your selection (for example, Fast ~341 tokens).
What happens after Generate
Jakka analyzes your context and surfaces recommendations as cards. Each card shows the opportunity type, impact, effort, and a proposed fix. Read more in the next article on the 5 recommendation types.