New Features

Configuration Reload

hister reload applies configuration, rules, and logging changes without a
manual server restart. It waits for completion and rejects invalid settings.
Expect a brief interruption and refresh open web pages afterward. Client
configuration changes still require restarting the affected clients.

More Expressive Search

Combine nested groups, phrases, and field filters with AND, OR, and NOT.
For example, (site:github.com OR site:gitlab.com) NOT title:archived searches
both sites while excluding archived titles. The new site: filter includes
subdomains, and has: finds documents with a populated field or metadata value,
such as has:label or has:metadata.author.

Invalid queries now explain what needs fixing and where. URL regular expression
searches are faster when a pattern has a specific starting prefix.

Kiwix, Sitemap, and Readwise Reader Imports

hister import zim collection.zim makes HTML and plain text from Kiwix archives
searchable without downloading each original page. It recovers original URLs
and article links, with --base-url as a fallback when archive paths match the
source website. Entries without an identifiable original URL are skipped with
a warning. Embedded images and other assets are not imported, and large
archives can require substantial memory.

hister import sitemap FILE_OR_URL indexes pages listed in XML sitemaps,
including sitemap indexes, gzip compressed files, and standard input. Duplicate
URLs are queued once, and the page crawl can be interrupted and resumed using
its job ID. Existing documents are skipped unless --force is supplied.

hister import readwise imports Readwise Reader documents with saved content,
notes, summaries, tags, and source metadata. It uses Reader's stored HTML when
available and otherwise tries the original page. Separate highlight and note
records are skipped, and original PDF and EPUB attachments are not downloaded.

Language Controls and Persistent Corrections

Limit automatic language detection with indexer.languages, or choose between
the default low memory mode and high accuracy mode with
indexer.language_detection_accuracy. Short web pages can use their titles to
help identify the language, and matching page declarations can supply regional
labels.

hister update QUERY --language en now pins the language across reindexing,
imports, and document updates. Use --language auto to clear a pin or
--language unknown to keep a document without an identified language.
language:en includes regional variants such as en-US, while a regional filter
narrows the results. Existing documents remain searchable when automatic
detection is disabled.

Expanded MCP Tools for AI Assistants

Assistants can page through search results, use the new get_facets tool to
explore document counts, and filter history by text and date. Search limits now
apply to the combined history, keyword, and semantic results.

Preview requests can select text, HTML, or metadata and read long documents in
smaller portions, reducing how much content an assistant needs to retrieve at
once.

Remote User Management and Proxy Authentication

The new hister user create, show, update, and delete commands manage
accounts on the configured running server using an administrator's token.
Account changes are applied together, conflicting updates are rejected, and
the last administrator cannot be deleted or demoted. Use
hister user create USERNAME --offline --admin with the server stopped to create
the first administrator or recover access.

server.proxy_auth_header enables sign in through a trusted authentication
proxy. The proxy must verify users, replace client supplied identity headers,
and prevent direct access to Hister. New proxy users receive ordinary accounts
without passwords. OIDC providers can also select a stable account identity
with identity_claim, including providers that do not supply email addresses.

Enhancements

1. Search interface: simpler result controls, clearer domains and titles,
   compact expandable snippets, and mobile friendly actions make results easier
   to scan. Set app.snippet_length to control the returned context and use
   Always expand snippets to remember a browser preference. Suggestions are
   hidden by default and can be enabled through Actions → Show suggestions.
2. Semantic search responsiveness: query embedding requests have separate
   capacity from indexing and a configurable timeout. Searches return keyword
   results when query embedding times out, with a default limit of two seconds.
3. PDF extraction: the built in reader handles more LaTeX generated PDFs.
   Configure indexer.pdf_parser to use an external parser such as pdftotext
   for browser uploads, watched files, and file imports.
4. Browser capture: the extension uses canonical URLs on the same hostname
   to reduce duplicate pages. Automatic capture skips HTTP error pages, and
   PDFs are recognized by their response type even without a .pdf suffix.
5. Previews and login: GitHub issue previews include discussion content,
   failed previews offer a fallback, and signing in returns to the original
   page with its query or preview link.
6. Feedback: maintenance commands show progress, command line searches
   explain empty results and can include document text, and the Add page
   explains submission failures.
7. Monitoring: server.expose_version enables public version reporting for
   monitoring clients and is disabled by default. Prometheus latency metrics
   use native histograms.
8. Dependencies: Go modules, npm packages, container bases, Nix inputs, and
   GitHub Actions were updated.

Bug Fixes

1. Anonymous searches in public mode return only global documents. Submitted
   documents cannot reference another document's stored HTML or favicon, and
   client redirects no longer leak access tokens to other destinations.
2. Password changes invalidate the affected user's existing browser sessions.
   GitHub login uses permanent account IDs so renamed or recycled usernames
   cannot change which identity an account belongs to.
3. Persistent crawls mark URLs complete only after successful document
   submission. Failed submissions are recorded as failures, and interrupted
   submissions remain resumable.
4. Cleanup preserves HTML and favicons while documents are still being indexed.
   Reindexing keeps files whose extracted text is empty, and exports preserve
   existing backups when writing fails.
5. Queries containing only negated terms return matching documents. Relevance
   optimization no longer broadens the requested results, and semantic results
   honor date filters and requested content fields.
6. Server access to local files is restricted to configured directories,
   including checks on symbolic link targets. Org files cannot read other files
   through include directives. Request, download, and extraction limits bound
   resource use.
7. PDF capture preserves pending downloads and uses the committed page title.
   The Firefox crawler rejects reported HTTP error responses, and qutebrowser
   handles larger browser events.
8. PostgreSQL timestamps are handled correctly, priority rule errors are
   reported, and repeated index write failures no longer cause excessive CPU
   use.

Backward Compatibility Notes

The following changes require attention when upgrading from v0.20.0:

1. GitHub and OIDC accounts: existing GitHub account links are not
   automatically migrated to permanent account IDs. Back up the database and
   arrange verified account relinking before users sign in again; otherwise,
   login creates a separate account while documents remain with the old one.
   Changing OIDC's identity_claim also requires relinking existing accounts.
   Leaving it unset preserves the existing email based identity.
2. User management scripts: the old create-user, show-user,
   update-user, and delete-user names remain deprecated aliases, but account
   operations now target the running server and require an administrator's
   personal token. Use user create --offline for local account creation with
   the server stopped. Replace the old token display flag with --show-token;
   --token now supplies authentication credentials.
3. Search syntax: uppercase AND, OR, and NOT are operators. Quote them
   to search for the words themselves. Negation takes precedence over AND,
   then OR; use parentheses to make grouping explicit. Malformed queries are
   rejected by search, update, and delete operations.
4. Local files: run hister reindex to remove previously indexed files
   reached through symbolic links outside configured directories. Org files
   using #+INCLUDE or #+SETUPFILE must contain the needed text directly.
   Opening a watched local file downloads it; use its preview to read it in
   Hister.
5. Size limits: document, PDF, and batch uploads default to 40 MiB, controlled
   by server.max_batch_body_size. HTTP HTML downloads and extracted text have
   a 32 MiB ceiling. Favicon and robots files have an 8 MiB limit; an oversized
   robots file causes the site's URLs to be skipped. Raising the upload limit
   does not raise extraction limits.
6. MCP clients: requests carrying an unrelated Origin are rejected. Set
   server.base_url to the public URL when using a reverse proxy. Clients
   paging opened history should pass both last_id and last_updated_at from
   the previous response.
7. Language settings: reindex to apply detection changes to existing
   documents. Reapply language corrections made before persistent pins were
   available if they should survive future updates.