New Features

Mastodon Extractor

A dedicated extractor for Mastodon detects Mastodon pages and indexes each
toot as its own separate document rather than one big blob of text. Works
with any Mastodon instance without configuration. Every toot gets a
metadata.type:toot field so you can filter toots in search queries.
Combined with a search alias (!toot → metadata.type:toot) this makes
finding past toots fast and convenient.

Metadata Query Filtering

Documents can now be filtered by arbitrary metadata fields using the
metadata.key:value query syntax. Extractors (including the newMastodon
extractor) populate these fields at index time.

Full Screen Preview

The split-screen preview can now be toggled into a full-screen mode that
occupies the entire content area (closes #401). The URL changes to
/preview/[id] so the view survives a page reload. Pressing the "view
result content" hotkey switches between split and full-screen. Full-screen
preview is also available on the history page.

Preview Panel on History Page

The history page now shows the same interactive preview panel as the search
page (closes #395). Hotkey navigation between history entries works the same
way as on the search page.

Infinite Scroll

Search results now load more pages automatically as you scroll to the bottom
of the list, removing the need to manually page through results (closes #1).

Image Lightbox

Images displayed inside the preview panel can be clicked to open a full-size
lightbox view.

Delete All Results

A new "delete all" action removes every document that matches the current
search query at once, without having to delete results one by one.

Keyboard Shortcut for Result Deletion

A dedicated hotkey deletes the currently focused search result directly from
the keyboard.

Quick Skip Rule

A skip-rule component lets you add a URL to your skip list directly from a
search result, available in both the web app and the browser extension
(closes #380).

robots.txt Support

The crawler and hister index now respect robots.txt by default (closes
#386). A new ignore_robots_txt config option disables this check when
needed.

Document Labeling

Documents can be tagged with custom labels when indexing from the CLI
(--label) or from the browser extension. Labels are stored as metadata
and can be used in search queries (closes #156, #373).

Editable Rules and Aliases

Existing skip/priority rules and search aliases can now be edited in the
web UI instead of having to delete and recreate them (closes #270).

Exact Phrase Matching

Multi-word queries now also attempt an exact phrase match across title and
text, so searching for open source surfaces pages that contain that exact
phrase more prominently (closes #394).

Extractor Templates and Extra Document Creation

Extractors can now supply a custom preview template, giving each content
type its own presentation in the preview panel. Extractors can also produce
additional sub-documents from a single page (used by the Mastodon extractor
to create one document per toot). A template scaffold is included to make
writing new extractors easier.

WebDriver BiDi Crawler Backend

A new bidi crawler backend uses the W3C WebDriver BiDi
protocol to drive an already-running browser over a WebSocket connection.
Unlike the chromedp backend, it does not launch a browser process: you start
the browser yourself (headless or not) and point Hister at it:

# Firefox
firefox --remote-debugging-port 9222

# Chrome / Chromium
chromium --remote-debugging-port=9222

crawler:
  backend: bidi
  backend_options:
    host: '127.0.0.1'
    port: '9222'
    capture_delay: 1.5 # extra seconds to wait after load for JS rendering

Supported by Firefox (≥ 102), Chrome/Chromium (≥ 106), and Edge. Options:
socket (full WebSocket URL), host, port, capture_delay. The crawler
reuses a single BiDi session for all URLs in one hister index run, making
multi-URL indexing significantly more efficient than opening a new browser
session per URL (closes #284).

Enhancements

• Resizable preview panel: drag the divider in split-screen view to
  adjust the panel width; the chosen width persists across sessions
• History hotkey navigation: keyboard navigation between history entries
  on the history page
• Secondary date sorting: when search scores are equal, results are
  sorted by indexed date
• yt-dlp multi-language subtitles: configure which subtitle languages
  the yt-dlp extractor indexes
• MCP date filtering: MCP search requests can now be filtered by date
• Semantic search chunking: punctuation-based boundaries used for
  chunk splitting, improving relevance for sentence-level queries
• OIDC enhancements: userinfo endpoint is now configurable for
  providers without auto-discovery (#279); password login can be disabled
  when OAuth is the only configured auth method
• Clearer CLI error messages: client-side HTTP errors now explain the
  problem in plain words and suggest which flag to use (#400)
• Duplicate rule prevention: the server and UI both reject duplicate
  skip/priority rules and aliases (#399)
• Deletion error feedback: errors during document deletion are now
  surfaced in the UI
• Rules/aliases UX: input fields moved above their respective lists
• Version string in server log: the server start message now includes
  the version number (#372)
• Silent WebSocket disconnect: closing the browser tab no longer shows
  a connection-error message

Bug Fixes

• Search terms are now properly escaped before query execution
• Focused result index computed correctly when priority results are present
• Phrase queries are only applied when no field-specific terms exist in the
  query
• /api/delete requests from the browser extension are now accepted
• text and html fields always included in search results (#374)
• Language field included in document search results
• OIDC scopes correctly forwarded to the provider (#371)
• Auth page is scrollable on small screens (#370)
• TUI client now passes the access token to the server (#368)
• Optional peer dependencies no longer excluded on non-Linux platforms (#299)