New work and a lot of tightening on top of v0.3.1 and v0.3.2. Your library now sorts the way you want, search results tell you what you already own, and the background full-text indexing that landed in v0.3.2 has been made steadier. The CLI and MCP server have caught up on the work they were missing.
If you drive linXiv from the CLI or an MCP client, read the Breaking changes section below. Several output shapes have moved this time.
Sort your library
The library was locked to newest-first. It now sorts by when a paper came out, when you added it, or its title, either way round, with a fitting starting order for each: in the app, on the CLI (linxiv list --sort --dir), and through the MCP list_papers tool. Your choice sticks between runs.
Search knows what you already have
Every search result used to come up unchecked, even for papers already sitting
in your library, so re-saving one was a single click and a wasted fetch. Both
arXiv and OpenAlex searches now report what you own: those rows come up checked
off, with the checkbox greyed out.
Searching for punctuation no longer throws an error either. It searches for it.
Steadier full-text indexing
The background indexer from v0.3.2 keeps working through your library, and now
handles a bad run better:
- A paper that fails is tried again a few times before being set aside, and
papers with no arXiv source are weeded out before the worker reaches for
them, so the backlog is a count of real work. - Repeated failures back off instead of hammering arXiv, and the idle path no
longer spins when there is nothing to do. - The backlog count is worked out in the database rather than by walking rows,
so a big library doesn't pay for the read-out. - Listing your library no longer drags the full text along with it, which is
the read that got slower as indexing filled in.
More sources
Crossref search now stands alongside OpenAlex as a full source, and
Crossref calls join the polite pool: the CROSSREF_MAILTO setting really does
reach Crossref now, the way OPENALEX_MAILTO already reached OpenAlex, which
buys you faster and steadier service.
CLI and MCP have caught up
Both gained the work they were missing: merging authors, listing and deleting
PDFs, looking up DOI candidates, adding in bulk, backlog and backup commands,
and weeding out duplicates.
Sharing
- Waiting invites now show up on the Share page, and a failed invite is rolled
back instead of leaving half a share behind. - Syncing reports counts and says more clearly what went wrong.
- Shares whose keys have drifted apart can be re-keyed.
Fixes
- Saving a PDF from arXiv first would fail. The PDF folder was never made
unless some other path made it first. - PDFs downloaded through the CLI didn't show up as saved.
- Papers brought in by DOI couldn't be reached. BibTeX wrote bare ids while
the CLI took everything for an arXiv id. You can now reach a paper by
whatever id you have. - Deleting a tag that was already on a paper or project now works.
- Every restore snapshot is kept, instead of each one writing over the last.
- Trash listings carry whole rows again.
- Editing files in a folder opened from disk works. The app was missing the
file-system rights. - Swapped out several unsafe dependencies.
Under the hood
A big tidy-up inside: wire types are now built from the Rust structs instead of
copied by hand, each paper, project and note shape has one place that writes
it, and callers no longer reach past the service layer into storage.
The search index is now worked out by the database rather than kept up by
hand. One view says what a paper can be searched by, and triggers keep the
index equal to it, so no write path can quietly let search drift away from your
library. Alongside that, the CLI's frozen output goldens are now really checked
by the test suite, having drifted unseen until this release, the saved-settings
upgrades are covered for every version step, and the whole workspace is
lint-clean.
Breaking changes (CLI and MCP)
Output shapes and error behaviour that scripts may lean on:
import_bibtexsource ids now carry a namespace.- CLI
listdropscreated_at/updated_at. CLI and MCPsearchnow send
SearchResultOutkeys (ADR-0011). - CLI and MCP exports move from published-DESC to the library ordering over
latest rows (ADR-0011). - CLI and MCP
statsgainrecent_papers. - The bulk-save wrapper is now
{saved_count, source_ids}, and CLI author
formatting follows core. - Adding or removing a paper from a project hands back a receipt. The CLI drops
its own one-off wrappers. - MCP PDF and full-text keys:
paper_idbecomessource_id. - MCP
get_project, and a missing note, now raise an error instead of handing
back JSONnull. - MCP Crossref reports an upstream outage as an error, not an empty list.
- 404 details now carry the id. MCP drops its trailing full stop, and CLI/notes
drop pyrepr quotes. - The CLI skip reason words the flag as
force=true.
All merged pull requests
What's Changed
- docs: seed graph randomness for reproducibility by @jakeuribe in #179
- Full text search by @jakeuribe in #178
- fixes: p2p syncing failures on new ticket schema by @jakeuribe in #182
- feat(FTS): background worker who works on fetching/indexing the full tex sources of arxiv papers in background by @jakeuribe in #181
- Updated readme to include information about python package by @jakeuribe in #185
- Dev by @jakeuribe in #186
- feat: add sort options and indixes on databases in library; fix: search behaves correctly for papers already saved to database by @jakeuribe in #184
- fix:resolved vulnerable dependencies by @jakeuribe in #183
- dependabot-to-dev by @jakeuribe in #188
- Chore/dependabot config by @jakeuribe in #187
- Dev by @jakeuribe in #189
- chore(deps): bump actions/checkout from 4 to 7 by @dependabot[bot] in #191
- chore(deps): bump actions/setup-node from 4 to 7 by @dependabot[bot] in #190
- Dependabot fixes by @jakeuribe in #192
- ci: restrict GITHUB_TOKEN to contents:read by @jakeuribe in #193
- unify layers by @jakeuribe in #194
- fix continuous integration by @jakeuribe in #195
- repo cleanup: docs, workflows etc... dev -> main by @jakeuribe in #196
- Cli mcp api by @jakeuribe in #197
- fix(pdf-storage): pdf directory would never be created if first pdf saved was through arxiv by @jakeuribe in #198
- docs: add Arch Linux build steps by @SparshSriva in #199
- merge conflicts by @jakeuribe in #202
- dev by @jakeuribe in #201
- refactor(service): recreating, same accessible seams for CLI, MCP, GUI, and anything else, to talk to. by @jakeuribe in #203
- Mc by @jakeuribe in #205
- Service seams by @jakeuribe in #204
- core hardening by @jakeuribe in #206
- unguarded invariants by @jakeuribe in #207
- feat: full text indexing, search, more by @jakeuribe in #208
- release asset order by @jakeuribe in #209
- docs: release asset ordering by @jakeuribe in #210
- ci(release): prefix raw binaries so they sort last by @jakeuribe in #211
- ci: lexicographic reorder of binary assets for PyPi by @jakeuribe in #212
New Contributors
- @dependabot[bot] made their first contribution in #191
Full Changelog: v0.3.2...v0.4.0
Release Update
macOS: On macOS 15 (Sequoia) and later, Apple removed the right-click → Open Gatekeeper bypass. The first launch will say "linXiv is da maged and can't be opened", this is not a broken download, it's Gatekeeper blocking an unsigned app. To open it:
- Try launching the app once (it will fail with the "damaged" message), then go to System Settings → Privacy & Security and click Open Anyway next to the linXiv block notice, or
- Run
xattr -cr /Applications/linXiv.appin a terminal to strip the quarantine attribute, then launch normally.
On macOS 14 and earlier, right-click the app → Open still works to get past Gatekeeper the first time.
Windows: click More info → Run anyway on the SmartScreen prompt.