oh-my-rss

KO | EN

gitlite — search

#pdf #computer-vision #postgresql #gemini #large-language-models #markdown #anthropic #developer-tools #pytorch #kotlin

(★ 9)

RSS-driven reading automation for FreshRSS, arXiv, Codex summaries, and static pages.

파일 탐색기

ZIP 다운로드

- - release.yml
  - test.yml
- github-cover-prompt.md
- github-cover-source.png
- github-cover.png
- example.yaml
- feed-management.md
- reeder-workflow.md
- synology-freshrss.md
- install-synology.sh
- make-github-cover.py
- nas_report_workspace.py
- - __init__.py
  - analytics.py
  - app.py
  - arxiv.py
  - arxiv_discovery.py
  - cli.py
  - codex.py
  - config.py
  - db.py
  - diagnostics.py
  - domains.py
  - feed_validation.py
  - freshrss.py
  - pdf.py
  - prompt.py
  - publisher.py
  - render.py
  - reports.py
  - scheduling.py
  - source_health.py
  - starter_opml.py
  - state.py
  - validation.py
- test_analytics.py
- test_app_discovery.py
- test_arxiv.py
- test_arxiv_discovery.py
- test_cli.py
- test_config.py
- test_db.py
- test_diagnostics.py
- test_domains.py
- test_feed_validation.py
- test_freshrss.py
- test_pdf.py
- test_project_metadata.py
- test_prompt.py
- test_publisher.py
- test_render.py
- test_scheduling.py
- test_source_health.py
- test_starter_opml.py
- test_state.py
- test_validation.py
.dockerignore
.env.example
.gitignore
CHANGELOG.md
docker-compose.yml
Dockerfile
LICENSE
MANIFEST.in
pyproject.toml
README.md
README.zh-CN.md

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

# 프로젝트 배지

Oh My RSS GitHub cover

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

[English](README.md) | [简体中文](README.zh-CN.md)

RSS-native AI research radar. Oh My RSS turns research feeds from FreshRSS into
Chinese paper-story summaries, static public feeds, category feeds, and monthly
trend reports that RSS clients such as Reeder can subscribe to.

## Features

- Reads RSS entries from a FreshRSS SQLite database.
- Optionally supplements subscribed feeds with broad arXiv API discovery so
  relevant papers outside subscribed arXiv subjects can still enter the queue.
- Detects arXiv, DOI, and generic paper links from RSS titles, links, and
  content.
- De-duplicates papers that appear in multiple feeds.
- Downloads direct PDF links when available and extracts text with `pdftotext`;
  otherwise falls back to the RSS abstract.
- Renders a first-page PNG preview and embeds it in the Codex summary page.
- Classifies papers into research-domain labels such as robot learning,
  manipulation, humanoids, VLA, navigation, SLAM, perception, safety/control,
  embodied AI, and benchmarks.
- Calls Codex CLI to generate Chinese summaries with:
  - `Motivation`
  - `Contribution`
  - `技术原理`
  - `实验设计及分析`
  - `原文链接`
- Produces static HTML with MathJax support.
- Produces a static public RSS feed at `feed.xml`.
- Exposes RSS and OPML auto-discovery links from the public index page.
- Produces a monthly research radar feed with trend tables and SVG charts.
- Produces a trending-topic feed with one item per hot research direction.
- Produces a trending-keyword feed for specific terms such as VLA, diffusion
  policy, humanoid, SLAM, safety filter, and sim-to-real.
- Produces a source health radar that records candidate counts, fetch failures,
  sudden zero-count drops, and stale venue years for each discovery source.
- Uses hash-based detail URLs to avoid stale browser/RSS-client caches.
- Optionally backs up the FreshRSS DB and updates entries with a clickable summary link.
- Prints a starter OPML bundle for common paper feeds so a new FreshRSS
  deployment can start quickly.
- Validates OPML feed URLs before importing them into FreshRSS.

## Project Goal

Oh My RSS is designed to become a self-hosted AI research radar rather than a
replacement RSS reader. The long-term goal is to let users connect paper feeds,
conference feeds, journal feeds, lab blogs, and news feeds, then publish a clean
knowledge stream with AI summaries, research-domain classification, trend
reports, and RSS-native distribution.

## Requirements

- Python 3.11+
- FreshRSS using SQLite
- `curl`
- `pdftotext` from Poppler
- PyMuPDF, installed automatically as a Python dependency
- Codex CLI authenticated on the machine running the job

## Quick Start

```bash
git clone https://github.com/LiuWhale/oh-my-rss.git
cd oh-my-rss
python3.11 -m venv .venv
. .venv/bin/activate
pip install -e ".[dev]"
oh-my-rss init-config --output config.yaml
```

Edit `config.yaml`, then run:

```bash
oh-my-rss doctor --config config.yaml
```

If the environment checks pass, generate summaries:

```bash
oh-my-rss run --config config.yaml --limit 1
```

To preview which papers would be processed without calling Codex:

```bash
oh-my-rss run --config config.yaml --dry-run --limit 5
```

To validate generated public site files before publishing or after a run:

```bash
oh-my-rss validate-site --site-dir ./site
```

To generate a starter OPML file for FreshRSS paper subscriptions:

```bash
oh-my-rss print-starter-opml --output starter-paper-feeds.opml
```

Validate it before importing:

```bash
oh-my-rss validate-opml --opml starter-paper-feeds.opml
```

Then import that OPML into FreshRSS and set `freshrss.category` to the imported
paper category.

## Configuration

See [`configs/example.yaml`](configs/example.yaml).

The key fields are:

- `freshrss.db_path`: path to FreshRSS `db.sqlite`.
- `freshrss.category`: FreshRSS category to scan, for example `论文` or `Papers`.
- `site.output_dir`: local directory where HTML files are written.
- `site.public_base_url`: public URL prefix for generated pages.
- `codex.command`: command used to invoke Codex CLI.
- `arxiv_discovery.enabled`: optional broad arXiv API discovery. When enabled,
  Oh My RSS fetches keyword-based arXiv candidates and classifies them by
  title/abstract content rather than exposing arXiv subject names as final
  public categories.
- `arxiv_discovery.max_results`: maximum arXiv API candidates to inspect per
  run.
- `arxiv_discovery.keywords`: optional replacement keyword list. Leave empty to
  use the built-in robotics, embodied AI, VLA, navigation, SLAM, tactile, and
  control-barrier discovery terms.
- `runtime.state_dir`: state, PDF cache, prompts, logs, and DB backups.

## Public Feed

Each run writes:

- `index.html`: public summary index
- `feed.xml`: public RSS feed for generated summaries
- `feeds.json`: machine-readable directory of all public RSS and OPML entry
  points
- `status.json`: machine-readable service status summary with summary counts,
  category counts, report counts, latest summary, and public feed URLs
- `robots.txt` and `sitemap.xml`: crawler discovery files for the public index,
  generated summary pages, monthly reports, hot directions, and hot keywords
- `opml.xml`: complete OPML import file for the main feed, category feeds,
  monthly report feed, hot direction feed, and hot keyword feed
- `categories/*.xml`: per-source/category RSS feeds
- `categories/index.json`: machine-readable category feed list
- `categories/opml.xml`: category-only OPML import file for RSS clients
- `reports/monthly.xml`: monthly research trend report RSS feed
- `reports/monthly/YYYY-MM.html`: monthly report pages with direction bars,
  source distribution, animated trend charts, and a filterable list of every
  paper in that month
- `reports/trending.xml`: hot research-direction RSS feed
- `reports/trending/index.html`: one shared hot-direction page with
  same-page direction filters and direct Chinese-summary links
- `reports/keywords.xml`: hot research-keyword RSS feed
- `reports/keywords/index.html`: one shared keyword-trend page with same-page
  keyword filters and direct Chinese-summary links
- `reports/source-health.xml`: source health RSS feed
- `reports/source-health/index.html`: source health radar page with per-source
  counts and warnings
- `reports/source-health/index.json`: machine-readable source health report
- `manifest.json`: machine-readable summary metadata

Category feed names are normalized for mixed paper sources: a leading `arXiv `
prefix is removed before publishing, and stale `categories/arxiv-*.xml` files
from older runs are cleaned up.

Newly generated records include `paper_id`, `source_kind`, and
`research_domains`. `source_kind` is `arXiv`, `DOI`, or `RSS`. Category feeds
and monthly reports use research-domain labels first, then fall back to
normalized feed names only when no research topic can be inferred.

Users can subscribe to:

```text
<site.public_base_url>/feed.xml
```

They can also subscribe to category-specific feeds. RSS clients generally do
not subscribe to JSON directly; use the complete OPML bundle for one-click
import:

```text
<site.public_base_url>/opml.xml
```

Use the category-only OPML file when you only want the topic/source feeds:

```text
<site.public_base_url>/categories/opml.xml
```

Use the JSON file only for integrations that need a machine-readable list:

```text
<site.public_base_url>/categories/index.json
```

For integrations that need every public RSS and OPML entry point, use:

```text
<site.public_base_url>/feeds.json
```

For monitoring or lightweight health checks, use:

```text
<site.public_base_url>/status.json
```

For crawler discovery, each run also writes:

```text
<site.public_base_url>/robots.txt
<site.public_base_url>/sitemap.xml
```

Monthly trend reports are published as a separate RSS feed:

```text
<site.public_base_url>/reports/monthly.xml
```

Each monthly report page includes an animated SVG trend chart, direction bar
chart, source distribution chart, and a static filterable paper list. Search,
source, and domain filters update the page URL and link back to the underlying
Codex paper summaries.

Hot research directions are also published as their own feed:

```text
<site.public_base_url>/reports/trending.xml
```

Each trending-topic item keeps its existing deep link into one shared direction
page. The matching direction is preselected there, alongside source filters,
text search, all matching papers, and links back to their generated summaries.

Specific research keywords are published as another RSS feed:

```text
<site.public_base_url>/reports/keywords.xml
```

Each keyword item keeps its existing deep link into one shared keyword page.
The matching term is preselected there, alongside source filters, text search,
all matching papers, and links back to their generated summaries.

Source health checks are published as an operational radar feed:

```text
<site.public_base_url>/reports/source-health.xml
```

The source health page and JSON report show which discovery sources produced
new candidates, which sources failed to fetch, which sources suddenly dropped
to zero candidates after a nonzero run, and which venue-year configurations
look stale:

```text
<site.public_base_url>/reports/source-health/index.html
<site.public_base_url>/reports/source-health/index.json
```

These feeds are static. They let other people read generated summaries without
logging into your FreshRSS account or sharing your read/unread state.

## Scheduling

Generate a locked cron entry for a 10-minute scheduler:

```bash
oh-my-rss print-cron \
  --cwd /opt/oh-my-rss \
  --config config.yaml \
  --limit 1 \
  --interval-minutes 10 \
  --log-path state/cron.log \
  --venv .venv
```

The command prints a cron line like:

```cron
*/10 * * * * cd /opt/oh-my-rss && . .venv/bin/activate && flock -n /tmp/oh-my-rss.lock oh-my-rss run --config config.yaml --limit 1 >> state/cron.log 2>&1
```

Paste that line into cron or the equivalent scheduler. Use `--no-venv` if
`oh-my-rss` is installed on the scheduler's default `PATH`.

## Deployment Notes

- For Synology NAS and FreshRSS Docker setups, see [`docs/synology-freshrss.md`](docs/synology-freshrss.md).
- For adding RSS subscriptions, grouping feeds, and OPML import, see [`docs/feed-management.md`](docs/feed-management.md).
- For Reeder/FreshRSS behavior, see [`docs/reeder-workflow.md`](docs/reeder-workflow.md).

For Docker Compose, copy `.env.example` to `.env` only when you want to override
the default run settings. The compose file has built-in defaults and does not
require a local `.env` file.

## Development

```bash
PYTHONPATH=src pytest -q
ruff check .
```

## Security

Do not commit:

- FreshRSS DB files
- Codex auth files
- real domains, private IPs, proxy credentials, or user accounts
- generated PDF caches

Use `.env.example` and `configs/example.yaml` as templates.

## License

MIT

oh-my-rss

오너의 다른 레포

파일 탐색기

# CDN으로 사용하기

# 프로젝트 배지

Was this content helpful?

Recent Feedback

oh-my-rss

오너의 다른 레포

파일 탐색기

# CDN으로 사용하기

# 프로젝트 배지

Was this content helpful?

Recent Feedback

Download README