Premium PDF generation
Han AI generates publication-grade PDFs in six distinct visual styles. The renderer learns which styles you prefer over time, so its choices get closer to your taste with every document.
What it does
Renders a PDF from structured content using pdfkit, with inline markdown, drop caps, cover pages, and footer pagination.
| Field | Value |
|---|---|
| Schema names | generate_pdf, record_pdf_feedback |
| Powered by | pdfkit |
| Styles | apple_keynote, vogue, bloomberg, executive, editorial, minimal |
| Preference store | /var/hanai/state/pdf-preferences.jsonl |
The six styles
| Style | Feel |
|---|---|
apple_keynote | Clean, generous whitespace, large headings |
vogue | Magazine-grade editorial — drop caps, serif body |
bloomberg | Dense, data-forward, terminal-style |
executive | Sober, briefing-paper layout |
editorial | Long-form essay treatment |
minimal | Stripped back, body-text-first |
When Han AI uses it
- A summary should leave the chat and become a deliverable.
- A briefing, proposal, or report needs to land in someone’s inbox as a real document.
- A recurring report (weekly, monthly) needs consistent presentation.
Examples
- “Generate the property-review PDF in editorial style.”
- “Take this and turn it into a one-page executive brief.”
- “Same format as last month’s report.”
Self-learning preferences
Every time you accept or reject a PDF, record_pdf_feedback writes the signal to /var/hanai/state/pdf-preferences.jsonl. Subsequent generations weight your preferred styles higher, so over time Han AI defaults to what you actually like instead of asking each time.
Limits
- PDFs are generated, not designed from scratch — the six styles are the design system.
- Custom logo and brand colours require a
TODO: confirm per-tenant brand-asset capture step. - Very long documents render but pdfkit is a streaming pipeline, not a desktop layout engine — expect document, not magazine spread.
See also
- Email — for delivering generated PDFs
- Capability gaps & learning
- Capabilities overview