Academic RSS Reader

简体中文

Turn scattered academic RSS feeds into a local reading queue that you can continuously screen, translate, and prioritize.

Academic RSS Reader is a desktop Obsidian plugin for the first pass of academic literature screening. It collects RSS and Atom feeds into a local SQLite database, keeps reading decisions in five clear baskets, and learns from those decisions to prioritize unread papers. It does not require a Python sidecar.

The plugin focuses on the step between “new papers arrived” and “I know what deserves a closer read”. It is designed to reduce repetitive triage, not to replace a reference manager or a full-text database.

[!IMPORTANT] Current recommended version: 1.7.0. This version requires Obsidian 1.13.0 or later and is desktop-only. Its bundled runtime must also provide Node.js 22.16 or later, node:sqlite DatabaseSync, and the SQLite Backup API. Update Obsidian before installing or updating when possible.

Why use Academic RSS Reader

• One reading queue: bring journal, lab, publisher, and other RSS/Atom feeds into one reader.
• Feedback-driven prioritization: mark papers as interested, archived, hidden, or expired, then use those decisions to rank the unread queue.
• A small, understandable workflow: collect, scan, classify, and return to the papers worth reading.
• Local-first storage: subscriptions, papers, reading states, translations, recommendations, and analysis stay in a Vault directory you choose.
• Optional assistance: translate visible titles or ask a configured OpenAI-compatible service to review only the papers that remain pending.
• Recoverable and explainable: migrations create a protection backup first; recommendation results retain positive and negative keyword evidence, while subscription defaults and article-level journal metadata stay separate.

The core principle is simple: the plugin handles repetitive collection and sorting; you decide what deserves attention.

How it works

Choose a Vault data directory
        ↓
Add or import RSS/Atom feeds
        ↓
Update feeds and collect new papers
        ↓
Scan the unread basket and mark reading states
        ↓
Refresh local recommendations
        ↓
Optionally review pending papers with an LLM

Who it is for

Academic RSS Reader is a good fit if you:

• follow multiple journals, publishers, laboratories, or research topics through RSS;
• receive more new papers than you can inspect immediately;
• want a queue that reflects your own reading decisions instead of a generic popularity ranking;
• prefer to keep research data inside your local Obsidian Vault;
• want lightweight title translation without sending every paper to an AI service.

What it can do

Collect and organize feeds

• Add, edit, enable, disable, and delete feeds.
• Import OPML, XML, TXT, pasted feed lists, or one URL per line.
• Update enabled feeds automatically when the reader is first opened after an app launch. Feeds updated successfully within the previous hour are skipped with a visible count; manual updates always run.
• Use ETag, Last-Modified, HTTP 304, bounded concurrency, retry delays, timeouts, cancellation, health states, and automatic backoff to make repeated updates less noisy.
• Extract titles, authors, source names, years, DOIs, links, and abstracts when provided by the feed.
• Preserve stable identities, normalized titles, cross-feed associations, existing reading states, and layered duplicate checks.

Screen papers in five baskets

• Manage papers as Unread, Interested, Archived, Hidden, or Expired.
• Load long lists continuously in batches of 100 instead of rendering the whole database at once.
• Choose whether cards show the journal, authors, publication date, DOI, feed-provided text abstract, and graphical abstract. Titles, relevance, and actions always remain visible.
• Sort any basket by title, last-seen update time, journal, or relevance.
• Search the current basket with fuzzy keywords across titles, authors, journals, abstracts, DOI, and feed metadata.
• After marking the papers you want as interested, hide every remaining unread paper in one confirmed action and undo the full batch when needed.
• Keep the basket navigation and search controls visible while scrolling; the Back to top action stays at the lower-right and returns to the list end position when you reach the bottom.
• Open the original paper in the system browser.
• Undo the most recent status action during the current reader session.

Prioritize the unread queue

• Train a local TypeScript TF-IDF and logistic-regression model from your reading states.
• Use interested and archived papers as positive examples, and hidden and expired papers as negative examples.
• Show high-relevance, pending, and low-relevance tiers below a fixed three-line title area, using a compact status badge and two aligned lines for the strongest positive and negative keyword evidence.
• Use the same TF-IDF feature scale for training and formal scoring; cancelled background training cannot write stale results.
• Manage the learned keyword list by disabling or re-enabling individual terms.
• Refresh recommendations automatically after feed updates, while reusing an unchanged model and scoring only new or changed unread papers when possible.
• Optionally review only pending papers with a user-configured OpenAI-compatible LLM. The LLM does not run automatically during normal feed updates.

The local model needs at least two positive and two negative training papers. Until there is enough feedback, papers can remain unscored rather than receiving an arbitrary personalized judgment.

Translate and analyze

• Translate visible paper titles and prefetch the next eight titles while you scroll.
• Cache translations in the local database and keep the original title available at any time.
• Review overall reading states and compare interest rates across feeds.
• Keep the interface language tied to Obsidian while choosing the content translation target separately. The source language is detected automatically for mixed-language feeds.
• Choose among Simplified Chinese (zh-CN), Traditional Chinese (zh-TW), English (en), Japanese (ja), Korean (ko), French (fr), German (de), Spanish (es), Portuguese (pt), Italian (it), and Russian (ru) as the translation target; the settings options show each code to make the saved target explicit.

Data model and safety boundaries

• When journal display is enabled, each card shows one journal name: an article-level value refreshed from RSS takes priority, with the earliest associated feed default used only as a fallback.
• Card settings default to the v1.5.0 presentation: journal and graphical abstract visible, with authors, publication date, DOI, and text abstract hidden.
• Every card uses the same height derived from the globally enabled rows. Titles reserve three lines, authors one line, and text abstracts three lines; missing item values and image loading do not change individual card height.
• Turning off graphical abstracts prevents card image elements and their remote requests from being created. Images are still loaded lazily and hidden on failure when enabled.
• Title fragments delimited by $...$, $$...$$, \(...\), or \[...\] use Obsidian's native MathJax renderer; invalid fragments fall back to their original text.
• Article links accept only http: and https:. RSS/XML rejects DOCTYPE, invalid root structures, and responses over 10 MiB; only known tracking parameters are removed, while business query parameters are preserved.
• LLM endpoints must use HTTPS, or HTTP on localhost, 127.0.0.1, or ::1. Requests have a 30-second timeout and bounded retries.

Get started

1. Open Settings → Community plugins → Browse, search for Academic RSS Reader, and install and enable it.
2. Open Settings → Academic RSS Reader and choose a data directory inside the current Vault.
3. Select Create new database for a new setup, or Load database for an existing rss-reader.sqlite3. v1.3 and older databases are upgraded in place; a protection backup is created first and the original database is restored if migration fails.
4. Open the reader from the RSS ribbon icon, or press Cmd/Ctrl + P and run Open reader.
5. Add feeds or import an OPML/feed list, then run an update.
6. Mark papers as you review them. After you have enough examples, open Personalized recommendations and run Update keyword recommendations.
7. If you want an additional review pass, configure an OpenAI-compatible endpoint, model, and SecretStorage API key in settings, then run Review pending items with LLM explicitly.

If the community directory is unavailable, download only main.js, manifest.json, and styles.css from the matching GitHub Release and place them in:

<Vault>/.obsidian/plugins/academic-rss-reader/
├── main.js
├── manifest.json
└── styles.css

Do not install the source-code archive generated by GitHub. When upgrading manually from the former rss-reader plugin ID, close Obsidian, rename the old plugin directory to academic-rss-reader, replace the three release files, and keep data.json so the legacy LLM key can be migrated to SecretStorage on first load.

Data and privacy

Your runtime database is stored in the Vault-relative directory you select:

<Vault>/<selected data directory>/
├── rss-reader.sqlite3
└── backups/

Subscriptions, papers, reading states, translations, recommendations, and interest analysis are stored locally in SQLite. General settings stay in the plugin's data.json; the LLM API key is stored through Obsidian SecretStorage, while data.json stores only the selected secret entry name.

Academic RSS Reader does not require an account, run telemetry, or operate a developer relay server. Network requests are limited to services you use:

• RSS/Atom updates go to the feed URLs you configure.
• Title translation uses an unofficial Google web endpoint. When enabled, visible and prefetched titles are sent directly from your device; translations may be rate-limited, inaccurate, or unavailable and should not be used for formal citation. Requests time out, retry with bounded backoff, and stop after repeated failures; a failed task remains available for manual retry.
• An LLM request is made only when you explicitly test the connection or review pending papers. The request includes the paper information needed for the review and your optional research-interest description, and is sent to the endpoint you configure.

The plugin does not create or open a database during startup. Database creation and loading begin only after you choose a Vault-relative data directory and explicitly create or load it. Saves use validated temporary and previous snapshots, and the settings page can create protection backups and restore the latest valid backup. The desktop-native SQLite boundary is limited to the database file, WAL/SHM sidecars, temporary files, and backups; other Vault files continue to use Obsidian's DataAdapter boundary.

Compatibility and boundaries

• Desktop Obsidian only: macOS, Windows, and Linux desktop installations.
• Requires Obsidian 1.13.0 or later plus a bundled Node.js 22.16+ runtime exposing node:sqlite; unsupported runtimes block database loading with an upgrade prompt. There is no sql.js fallback.
• No Python, separate Node.js installation, or sidecar process is required; Obsidian supplies the required runtime.
• Title translation is an experimental reading aid, not a formal translation or citation service.
• The database is intended for one local desktop Obsidian instance at a time; it is not a multi-user sync database.

Continue reading

• Documentation index
• Architecture design
• Database design
• Development guide
• Troubleshooting
• Security and privacy
• Changelog
• v1.7.0 release notes
• v1.6.3 release notes
• v1.6.2 release notes
• v1.6.1 release notes
• v1.6.0 release notes
• v1.5.0 release notes
• v1.4.1 release notes
• v1.4.0 release notes
• v1.3.0 release notes
• v1.2.0 release notes

Development and release commands are documented in the development guide.

License

MIT