noisepan

name	noisepan
description	Local-only signal extractor for noisy information streams (RSS, Telegram, Hacker News)
user-invocable	false
metadata	{"requires":{"bins":["noisepan"]}}

noisepan — Signal Extraction

You have access to noisepan, a deterministic signal extractor for information streams. It reads posts from RSS, Telegram, and Hacker News, scores them against a keyword-based taste profile, and outputs a ranked digest. No ML, no cloud, all local.

Install

brew install ppiankov/tap/noisepan

Setup

noisepan --config ~/.noisepan init

This creates ~/.noisepan/config.yaml (sources) and ~/.noisepan/taste.yaml (scoring weights). Edit both before first use.

Minimal config.yaml (RSS only)

sources:
  rss:
    feeds:
      - "https://www.reddit.com/r/devops/.rss"
      - "https://feeds.feedburner.com/TheHackersNews"
      - "https://www.cisa.gov/cybersecurity-advisories/all.xml"

Minimal taste.yaml

weights:
  high_signal:
    "cve": 5
    "zero-day": 5
    "kubernetes": 3
  low_signal:
    "webinar": -4
    "hiring": -3

thresholds:
  read_now: 7
  skim: 3
  ignore: 0

Commands

Command	What it does
noisepan pull	Fetch new posts from all configured sources
noisepan digest	Score all posts, output ranked digest
noisepan digest --format json	Machine-readable JSON output
noisepan digest --format markdown	Markdown table output
noisepan digest --output /tmp/d.md --format markdown	Write digest to file
noisepan digest --webhook https://hooks.slack.com/...	POST JSON digest to webhook
noisepan run	Pull + digest in one step
noisepan run --every 30m	Continuous mode (pull + digest every 30m)
noisepan stats	Per-channel signal-to-noise ratios, scoring distribution
noisepan stats --since 7d	Stats for last 7 days
noisepan stats --format json	Machine-readable stats for scripted monitoring
noisepan rescore	Recompute all scores with current taste profile
noisepan rescore --since 7d	Rescore only last 7 days
noisepan verify	Check source credibility of read_now posts via entropia
noisepan import feeds.opml	Import RSS feeds from OPML file
noisepan import feeds.opml --dry-run	Preview what would be imported
noisepan explain <post-id>	Show scoring breakdown for one post
noisepan doctor	Health check: config, database, sources, feed health
noisepan version	Print version

Key Flags

Flag	Default	Description
--config DIR	.noisepan/	Config directory
--since DUR	24h / 30d	Time window (e.g., 48h, 7d)
--format FMT	terminal	Output: terminal, json, markdown (stats: terminal, json)
--source SRC	all	Filter by source type
--channel CH	all	Filter by channel name
--no-color	false	Strip ANSI colors
--output PATH	stdout	Write digest to file
--webhook URL	off	POST digest JSON to URL
--dry-run	false	Preview import without modifying config

Agent Usage Pattern

For programmatic use, always use --format json:

noisepan pull && noisepan digest --format json --since 24h

JSON Output Structure

{
  "meta": { "channels": 28, "total_posts": 191, "since": "1d" },
  "trending": [
    { "keyword": "CVE-2026-1234", "channels": ["CISA", "Krebs", "BleepingComputer"] }
  ],
  "read_now": [
    {
      "source": "rss",
      "channel": "CISA",
      "url": "https://www.cisa.gov/...",
      "posted_at": "2026-02-20T10:00:00Z",
      "score": 10,
      "tier": "read_now",
      "labels": ["critical"],
      "headline": "First sentence of the post..."
    }
  ],
  "skims": [],
  "ignored": 164
}

Parsing Examples

# Get all read_now items
noisepan digest --format json | jq '.read_now'

# Get URLs of critical posts
noisepan digest --format json | jq -r '.read_now[] | select(.labels | index("critical")) | .url'

# Get trending topics
noisepan digest --format json | jq '.trending'

# Save digest to file and post to webhook
noisepan digest --output ~/digest.json --format json --webhook https://hooks.slack.com/...

Typical Workflow

1. One-time setup: noisepan init → edit config.yaml + taste.yaml
2. Add feeds: Edit ~/.noisepan/config.yaml or noisepan import feeds.opml
3. Tune weights: Edit ~/.noisepan/taste.yaml to match the operator's interests
4. Daily digest: noisepan run (or cron: 0 8 * * * noisepan run --output /tmp/digest.json --format json)
5. Check analytics: noisepan stats to see signal-to-noise ratios per feed
6. Verify top posts: noisepan verify (requires entropia in PATH)
7. Automated delivery: noisepan run --webhook https://hooks.slack.com/...

Integration with entropia

The verify command calls entropia scan <url> --json on each read_now post. Install entropia first:

brew install ppiankov/tap/entropia

Posts from domains that can't be scanned (reddit.com, t.me) are skipped automatically.

What noisepan Does NOT Do

• Does not send messages or interact with sources
• Does not use ML or embeddings — deterministic keyword scoring only
• Does not sync to cloud — all data in local SQLite
• Does not run as a daemon — pull-based, runs when invoked
• Does not replace reading — reduces what needs to be read

Exit Codes

• 0 — success
• 1 — error (details on stderr)

---

This tool follows the Agent-Native CLI Convention. Validate with: ancc validate .