Metadata-Version: 2.5
Name: news-recap
Version: 1.10.0
Project-URL: Homepage, https://andgineer.github.io/news-recap/
Project-URL: Documentation, https://andgineer.github.io/news-recap/
Author-email: Andrey Sorokin <andrey@sorokin.engineer>
License: Copyright (c) 2026 Andrey Sorokin <andrey@sorokin.engineer>
        
        Permission is hereby granted, free of charge, to any person obtaining a copy of
        this software and associated documentation files (the “Software”), to deal in
        the Software without restriction, including without limitation the rights to use,
        copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the
        Software, and to permit persons to whom the Software is furnished to do so,
        subject to the following conditions:
        
        The above copyright notice and this permission notice shall be included in all
        copies or substantial portions of the Software.
        
        THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND,
        EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
        OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
        NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
        HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
        WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
        FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
        OTHER DEALINGS IN THE SOFTWARE.
License-File: LICENSE.txt
Keywords: one,two
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Classifier: Programming Language :: Python :: 3
Requires-Python: >=3.12
Requires-Dist: anthropic>=0.40
Requires-Dist: click>=8.2.0
Requires-Dist: defusedxml>=0.7.1
Requires-Dist: flask>=3.1
Requires-Dist: httpx>=0.28.0
Requires-Dist: langcodes[data]>=3.5.1
Requires-Dist: msgspec>=0.19
Requires-Dist: python-dotenv>=1.0
Requires-Dist: rich-click>=1.8.8
Requires-Dist: sentence-transformers>=3.0.1
Requires-Dist: tomlkit>=0.13
Requires-Dist: trafilatura>=2.0.0
Requires-Dist: typesafe-sdk>=0.7.2
Requires-Dist: youtube-transcript-api>=1.0.0
Provides-Extra: api-litellm
Requires-Dist: litellm>=1.40; extra == 'api-litellm'
Description-Content-Type: text/markdown

[![Build Status](https://github.com/andgineer/news-recap/workflows/CI/badge.svg)](https://github.com/andgineer/news-recap/actions)
[![Coverage](https://raw.githubusercontent.com/andgineer/news-recap/python-coverage-comment-action-data/badge.svg)](https://htmlpreview.github.io/?https://github.com/andgineer/news-recap/blob/python-coverage-comment-action-data/htmlcov/index.html)
# news-recap

**Your feeds. Fewer repeats. A clearer picture.**

Turn RSS feeds into a daily digest: related reports grouped together, vague
headlines clarified, and links back to the original sources. Choose the topics
to follow, the noise to exclude, and the language to read in.

* **Read stories, not a wall of headlines.** Related reports come together in
  topic sections, with summaries and links to their sources.
* **Get past the clickbait.** When a headline hides the news, the pipeline fetches
  the article and rewrites the headline from its content.
* **Have it ready when you are.** Schedule digest creation and read it in the
  browser. Interrupted runs resume from saved progress.

## Quick start

The free way needs no API keys: install [uv](https://docs.astral.sh/uv/getting-started/installation/)
and the [Antigravity CLI](https://antigravity.google/) (`agy`), then run:

```bash
uv tool install news-recap --upgrade --python 3.13
news-recap config set rss "YOUR_RSS_URL"
news-recap config set language en    # the digest language; default ru
news-recap ingest
news-recap create
news-recap serve
```

Settings live in `~/.news_recap_data/config.toml` (`news-recap config` shows them); use
`news-recap config set agent claude` or `codex` for a subscription agent. For scheduled runs,
optional Jev filtering and all settings, see the [manual](https://andgineer.github.io/news-recap/).

## Under the hood

**Storage that fits the workflow.** I started with SQLite, SQLModel, and Alembic,
then found that the database added more machinery than this batch-processing
workflow needed. Typed `msgspec` structures now carry data through processing
and JSON storage, with daily article partitions and resumable checkpoints.
The [database prototype](https://github.com/andgineer/news-recap/tree/sql-poc)
remains available for comparison.

**Spend model calls where judgment matters.** Embeddings narrow down candidate
duplicates; LLMs decide which reports belong together and write the digest.
Code removes duplicate and overlapping blocks after generation. Full articles
are fetched selectively, when their headlines need clarification.

**Coding agents as pipeline workers.** The pipeline invokes Codex, Claude Code,
or Antigravity through their CLI interfaces, with task-specific model selection
and saved intermediate results. The [agent integration guide](spec/agents.md)
describes the launch commands and task contracts for adapting this approach to
another batch-processing application.

### Development process

In this project, I experimented with an iterative development workflow using
Codex and Claude Code: written plans, implementation, and separate review passes,
followed by trials on real news feeds.

Those trials also shaped the architecture. The
[pipeline experiments](spec/pipeline.md#experiments) compare digest strategies
on the same article corpus, including processing time, article coverage, and
the resulting section structure.

<details>
<summary><b>Contributing</b></summary>

Bootstrap the development environment and install pre-commit hooks:

```bash
source ./activate.sh
pre-commit install
```

Run the checks and test suite:

```bash
uv run inv pre
uv run pytest --cov=src tests/
```

See the [contributor guide](AGENTS.md) for repository conventions and
`uv run invoke --list` for available development tasks.

[Allure test report](https://andgineer.github.io/news-recap/builds/tests/)

</details>
