GitOps Daily AI News Aggregator

A small, serverless daily AI news pipeline that runs entirely in GitHub:

1. GitHub Actions runs on a schedule.
2. Python fetches RSS/Atom feeds.
3. The workflow calls GitHub Models using the automatic GITHUB_TOKEN.
4. The script writes Markdown and JSON into the repository.
5. GitHub Actions commits the generated files back to main.
6. GitHub Pages can serve /docs as a static daily news site.

No database, server, newsletter provider, or separate LLM key is required for the default GitHub Actions setup.

What gets generated

digests/YYYY-MM-DD.md          # human-readable digest in the repo
docs/index.md                  # GitHub Pages homepage
docs/digests/YYYY-MM-DD.md     # GitHub Pages copy of the digest
data/items/YYYY-MM-DD.json     # selected source items and scores
data/seen_urls.json            # lightweight de-duplication state

Optional issue delivery is included: set repository variable CREATE_GITHUB_ISSUE=true and the workflow will also create one GitHub Issue per daily digest.

Quick start

1. Create a new GitHub repository. A public repository is recommended for the free GitHub-hosted runner setup.
2. Copy these files into the repository.
3. Commit and push to the default branch.
4. Open Actions and run Daily AI News GitOps manually once.
5. Open Settings → Pages and set Build and deployment → Source to GitHub Actions.
6. Check digests/, data/items/, and the Pages deployment after the workflow completes.

The workflow is scheduled for 5:17 AM in America/Vancouver. Change this in .github/workflows/daily-ai-news.yml.

Architecture and workflow

Detailed Workflow

GitHub Models setup

Inside GitHub Actions, no extra model secret is needed. The workflow grants:

permissions:
  contents: write
  models: read
  issues: write
  pages: write
  id-token: write

The script calls https://models.github.ai/inference/chat/completions with ${{ secrets.GITHUB_TOKEN }}.

For local AI testing, create a GitHub personal access token with the models scope and run:

export GITHUB_PAT="your-token"
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python src/ai_news_aggregator/main.py --config config/sources.yml --dry-run

For local non-AI testing:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python src/ai_news_aggregator/main.py --config config/sources.yml --no-ai --dry-run

Customize sources

Edit config/sources.yml.

Each feed supports:

- name: "Example AI Blog"
  url: "https://example.com/feed.xml"
  weight: 1.5
  entry_limit: 40

Scoring is intentionally simple and auditable:

• source weight
• recency
• keyword matches
• de-duplication by canonical URL and normalized title
• optional suppression of already-seen URLs for a configurable number of days

Useful repository variables

Set these in Settings → Secrets and variables → Actions → Variables:

Variable	Default	Purpose
CREATE_GITHUB_ISSUE	false	Create one GitHub Issue per digest so watchers can get notifications.

Set these only if you need to override the defaults:

Environment variable	Purpose
GITHUB_MODELS_MODEL	Override the model in config/sources.yml.
GITHUB_MODELS_ENDPOINT	Override the GitHub Models endpoint.

GitOps notes

This is GitOps in the practical sense: the repository contains the workflow, source configuration, generated artifacts, and state. Changes to sources or scoring are pull requests. Daily output is committed history. Rollback is a git revert.

The design works best for a public static digest. Do not commit private feeds, paid-content excerpts, API keys, cookies, or personal emails to the repository.

Limits to expect

• GitHub scheduled workflows are not real-time cron. Runs can be delayed and should be scheduled away from the top of the hour.
• Public-repository scheduled workflows can be disabled after long periods with no repository activity. Daily commits usually keep the repo active.
• GitHub Models free usage is rate-limited. This project uses one model call per run by default.
• Some websites block bots or do not provide RSS. Prefer official RSS/Atom feeds.
• GitHub Pages is static hosting; it will not run a backend or database. The workflow deploys Pages via actions/upload-pages-artifact and actions/deploy-pages, rather than relying on a separate branch-build trigger.

Project files

.github/workflows/daily-ai-news.yml      # daily scheduled GitHub Actions workflow
.github/prompts/daily_digest.prompt.yml  # reusable prompt for GitHub Models UI
config/sources.yml                       # feed list, scoring, model settings
src/ai_news_aggregator/main.py           # collector, scorer, summarizer, writer
requirements.txt                         # Python dependencies
docs/                                    # GitHub Pages output
digests/                                 # generated Markdown archive
data/                                    # generated JSON/state