v3.5.0 — UX polish

Small touches that make DocuLens feel faster and more alive: a progress bar and rotating document facts on the reading screen, instant starter questions that swap to tailored ones, a cleaner thinking panel, and a catchier landing with an example of a cited answer.

v3.4.0 — Calm Light (Claude-style) redesign

Adopts Claude's own design language (Claude Calm Light): ink text and black primary buttons on a warm near-white ground, a single terracotta accent used only for brand and emphasis, a light editorial serif (Newsreader) for headlines with Inter for the interface, hairline borders and soft flat depth. Calm, editorial, spacious — in both light and dark.

v3.3.0 — Clean Minimal redesign

A quieter, more professional look (Linear/Claude register): neutral surfaces, a single restrained blue accent, Inter throughout (no more decorative serif), smaller radii and lighter shadows, subtler motion. Also fixes the indexing spinner that sat frozen on machines with reduced-motion enabled.

v3.2.0 — Premium answers (rich Markdown)

Answers now render like Claude/ChatGPT: proper headings, bullet and numbered lists, tables, inline/fenced code, bold and italic — with page citations as inline chips. Rendered safely (no innerHTML). A big jump in reading experience over the old flat plain text.

v3.1.1 — Request-ID log fix

Fixes the per-request access-log line showing '-' instead of the request id (the contextvar was reset before the line was logged).

v3.1.0 — Observability

Production-hardening: structured logging with per-request IDs (X-Request-ID), each API call logged with status + latency, and optional Sentry error reporting (off unless SENTRY_DSN is set; never includes document content). Also re-tested the live site from a Kali box (nmap/nikto/sslscan) — no vulnerabilities found.

v3.0.0 — DocuLens

The project is now DocuLens — An AI Powered Document Assistant.

• Rebranded everything: UI, backend, docs, the repository, and the live domain → https://doculens.duckdns.org.
• Fixed a bug (v2.5.1) where uploading a second file with the same name returned 422 after the v2.5.0 multi-format change.

No behaviour change beyond the fix. 119 tests pass.

v2.5.0 — Word, text & Markdown support

The assistant now reads PDF, Word (.docx), plain text and Markdown, not only PDFs. Non-PDF files are split into even pages so page citations stay meaningful. (A sentence-aware chunker was trialled and dropped after it regressed the retrieval eval.)

v2.4.0 — Reading Room redesign

A design with a point of view, and a footer that behaves like one.

• New look ("Reading Room"): warm paper, a self-hosted Fraunces editorial serif for headlines and the wordmark, Inter for the interface, monospace numerals, one ink-green accent, and highlighter amber reserved for citations. Warm, lamp-lit dark mode.
• The footer is a real footer: it closes the opening page like a colophon (scroll to reach) instead of a pinned bar, and steps aside in a conversation — where the version and live provider health move to the top bar.
• Editorial motion: things settle with a soft spring, staggered, instead of a uniform fade-up.
• Mobile top bar tidied so the wordmark never wraps.

v2.3.1 — Mobile footer polish

Cleaner mobile footer: the credit sits left, the version pill is balanced at the top-right, and the links sit evenly on their own row (no more cramped version pill jammed against the links). Verified at 390/360/320px.