NOOZ

Your AI-powered HPC intelligence hub. Nooz scours the web, aggregates news from credible HPC sources, and delivers actionable intelligence about supercomputing systems, software, and trends—all in a clean, sourced report.

Why Nooz?

HPC administrators and engineers face information overload. Vendor blogs, academic papers, release notes, and news outlets produce a constant stream of updates. Missing a critical patch, a new best practice, or an emerging technology can have real consequences.

Nooz is your automated research assistant. It:

• Monitors RSS feeds from HPC news leaders (HPCwire, The Next Platform, Phoronix, LWN, AnandTech)
• Fetches full article content from across the web
• Curates findings to remove noise and duplicates
• Discovers new sources and emerging topics
• Sources every claim with proper citations and credibility ratings
• Evolves itself—after each run, it suggests new agents, sources, and improvements

What It Does

Every week (or on demand), Nooz generates a structured intelligence report covering:

• Major Themes — Deep dives into the most important HPC trends
• Important Updates — Version releases, security patches, feature announcements
• Emerging Topics — New technologies worth watching
• Source Credibility Summary — Know the provenance of your information

Every finding in the report links back to its source with a URL. Nooz marks credibility: [High Credibility] for peer-reviewed papers and official documentation, [Medium Credibility] for established tech journalism, [Verify] for unconfirmed claims.

Key Features

Multi-Agent Architecture

Agents are defined in simple Markdown files. Add a new scout by dropping a .md file into the agents/ directory. No code required.

Parallel Scout Execution

Scouts run concurrently, fetching from multiple RSS feeds simultaneously. Fast by default.

Source Verification

Nooz verifies that feeds are healthy and reachable before scouting. Dead sources are flagged.

Full-Content Fetching

Don't settle for RSS summaries. Nooz fetches the full article content for deeper analysis.

Self-Evolving

After each report, the evolution agent identifies gaps in coverage, suggests new agents and sources, and recommends prompt improvements. Use python evolve.py to apply those changes.

Quick Start

# Clone and set up
git clone https://github.com/your-org/nooz.git
cd nooz
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

# Configure
export OPENROUTER_API_KEY=sk-or-...

# Run
python run.py

Configuration

Environment Variables

Variable	Required	Description
OPENROUTER_API_KEY	Yes	API key for OpenRouter LLM access

Adding Sources

RSS feeds are configured in core/fetchers.py under RSS_FEEDS. Add new feeds by extending the dictionary:

RSS_FEEDS = {
    "myfeed": {
        "url": "https://example.com/feed/",
        "topics": ["hpc", "gpu"],
        "credibility": 0.8
    },
}

Adding Agents

Create a new agent in agents/ directory:

---
name: my-scout
role: HPC GPU specialist
model: openrouter/free
skills:
  - summarize
  - classify
sources:
  - rss/hpcwire
  - rss/nextplatform
output: workspace/findings/gpu.md
---

## Mission

Track GPU scheduler developments, CUDA releases, and GPU cluster management.

The Evolution Loop

Nooz is self-improving. After each run:

1. workspace/memory/evolution.md is generated with suggestions
2. Run python evolve.py --dry-run to preview changes
3. Run python evolve.py (or --auto) to apply changes
4. New agents, skills, and sources are created automatically

Architecture

run.py
 └── run_scouts()      → Fetch RSS, analyze articles
     └── run_agent()   → Per-agent processing
 └── curate_findings() → Deduplicate, filter, prioritize
 └── discover_topics() → Find new sources, identify gaps
 └── generate_report()→ Produce final intelligence report
 └── evolve_swarm()    → Generate evolution suggestions

Agents, skills, and sources are all Markdown-defined. The system is human-editable at every layer.

Built With

• OpenRouter — Unified LLM API
• feedparser — RSS parsing
• trafilatura — Full-text web extraction
• Python 3.10+ — Async-first architecture

License

MIT