News Briefings

Adjutant can deliver daily curated news briefings from Hacker News, Reddit, and RSS feeds. The pipeline fetches articles, ranks them with an LLM, and delivers a digest via Telegram and/or journal entry.

How It Works

The news pipeline has three stages:

1. Fetch -- pulls articles from configured sources (HN Algolia API, Reddit JSON API, RSS/HTML feeds) based on keywords
2. Analyze -- sends fetched articles to an LLM for relevance ranking and summarization
3. Deliver -- formats the top stories as a markdown briefing, sends via Telegram, and writes to the journal

Running a Briefing

Manual

adjutant news

Scheduled

Add a schedule in adjutant.yaml:

schedules:
  - name: "news_briefing"
    description: "Daily AI news briefing"
    schedule: "0 8 * * 1-5"  # Weekdays at 8am
    script: ".venv/bin/python -m adjutant news"
    enabled: true

Then sync the schedule to crontab:

adjutant schedule sync

Configuration

News settings live in news_config.json at the Adjutant root. An example template is provided at news_config.json.example.

Full Schema

{
  "keywords": ["AI agents", "LLM", "transformer"],

  "sources": {
    "hackernews": {
      "enabled": true,
      "max_items": 20,
      "lookback_hours": 24
    },
    "reddit": {
      "enabled": false,
      "subreddits": ["MachineLearning", "LocalLLaMA"],
      "max_items": 20,
      "lookback_hours": 24
    },
    "blogs": {
      "enabled": false,
      "feeds": [
        {"name": "Example Blog", "url": "https://example.com/feed", "type": "rss"}
      ]
    }
  },

  "analysis": {
    "model": "anthropic/claude-haiku-4-5",
    "top_n": 3,
    "prefilter_limit": 20
  },

  "deduplication": {
    "window_days": 30
  },

  "delivery": {
    "telegram": true,
    "journal": true
  },

  "cleanup": {
    "raw_retention_days": 7,
    "analyzed_retention_days": 7
  }
}

Key Settings

Setting	Description
keywords	Search terms used across all sources. Combined with OR logic.
sources.hackernews.enabled	Fetch from Hacker News Algolia API
sources.reddit.subreddits	Which subreddits to search
sources.blogs.feeds	RSS/Atom feed URLs to check
analysis.model	Which LLM to use for ranking (default: Haiku for cost)
analysis.top_n	Number of top stories to include in the briefing
delivery.telegram	Send the briefing via Telegram
delivery.journal	Write the briefing to the journal
deduplication.window_days	How long to remember seen articles to avoid repeats

Output

A typical briefing looks like:

Agentic AI News -- 16.03.2026

1. New Architecture for Long-Context Agents
   -> https://example.com/article1
   Novel approach to maintaining context over extended agent sessions...

2. OpenCode 2.0 Released
   -> https://opencode.ai/blog/v2
   Major update to the OpenCode runtime with improved tool calling...

3. Self-Improving Agents: A Survey
   -> https://arxiv.org/abs/2026.12345
   Comprehensive survey of autonomous agent self-improvement techniques...

Intermediate Files

The pipeline stores intermediate results in state/:

Path	Contents
state/news_raw/YYYY-MM-DD.json	Raw fetched articles
state/news_analyzed/YYYY-MM-DD.json	LLM-ranked results
state/news_dedup.json	Deduplication cache (seen URLs)

Old files are cleaned up automatically based on cleanup.*_retention_days.

Dependencies

• The RSS/blog source requires the optional feedparser package: pip install feedparser
• All sources require network access