Wachi

wachi

Subscribe any link and get notified on change. Monitors URLs for new content, pushes notifications to 90+ services via apprise.

Install

# npm / bun (no install needed)
npx wachi --help
bunx wachi --help

# global install
npm i -g wachi
bun i -g wachi

# shell script
curl -fsSL https://raw.githubusercontent.com/ysm-dev/wachi/main/install.sh | sh

# homebrew
brew tap ysm-dev/tap && brew install wachi

Quick Start

# 1. Subscribe to any URL (auto-discovers RSS)
wachi sub "slack://xoxb-token/channel" "https://blog.example.com"

# 2. Check for new content (run on a schedule)
wachi check

# New posts get pushed to Slack. That's it.

Commands

wachi sub <apprise-url> <url>     Subscribe URL to notification channel
  -e, --send-existing             Send all current items on next check (skip baseline)

wachi unsub <apprise-url> [url]   Unsubscribe URL or remove entire channel

wachi ls                          List all channels and subscriptions

wachi check                       Check all subscriptions for changes
  -c, --channel <apprise-url>     Check specific channel only
  -n, --concurrency <number>      Max concurrent checks (default: 10)
  -d, --dry-run                   Preview without sending or recording

wachi test <apprise-url>          Send test notification

wachi upgrade                     Update wachi to latest version

Global flags: --json / -j for machine-readable output, --verbose / -V for detailed logs, --config / -C for custom config path.

How It Works

1. wachi sub checks if the URL has an RSS feed (auto-discovery via <link> tags and common paths)
2. If RSS found: store and use RSS for ongoing checks
3. If no RSS: use LLM + agent-browser to identify CSS selectors via accessibility tree analysis
4. wachi check fetches each subscription, compares against dedup table (SHA-256 hash), sends new items via apprise

Configuration

Config at ~/.config/wachi/config.yml (auto-created on first wachi sub).

# LLM config (only needed for non-RSS sites)
# Also settable via WACHI_LLM_API_KEY, WACHI_LLM_MODEL env vars
llm:
  api_key: "sk-..."
  model: "gpt-4.1-mini"

# Optional: summarize articles before sending
summary:
  enabled: true
  language: "en"
  min_reading_time: 3  # minutes

# Channels managed by wachi sub/unsub
channels:
  - apprise_url: "slack://xoxb-token/channel"
    subscriptions:
      - url: "https://blog.example.com"
        rss_url: "https://blog.example.com/feed.xml"

All fields optional with sensible defaults. Empty config is valid.

Environment Variables

Notification Channels

Uses apprise URL format. Examples:

# Slack
wachi sub "slack://xoxb-token/channel" "https://example.com"

# Discord
wachi sub "discord://webhook-id/token" "https://example.com"

# Telegram
wachi sub "tgram://bot-token/chat-id" "https://example.com"

# Test channel works
wachi test "slack://xoxb-token/channel"

Full list: https://github.com/caronc/apprise/wiki

Scheduling

wachi check is stateless and one-shot. Use any scheduler:

# crnd (recommended)
crnd "*/5 * * * *" wachi check

# system cron
crontab -e
# */5 * * * * wachi check

Examples

# Blog (auto-discovers RSS)
wachi sub "slack://xoxb-token/channel" "https://blog.example.com"

# Hacker News (LLM identifies selectors)
wachi sub "discord://webhook-id/token" "https://news.ycombinator.com"

# YouTube channel
wachi sub "tgram://bot-token/chat-id" "https://youtube.com/@channel"

# URL without https:// (auto-prepended)
wachi sub "slack://token/channel" "blog.example.com"

# Send existing items on next check
wachi sub -e "discord://webhook-id/token" "https://news.ycombinator.com"

# Dry-run check
wachi check -d

# Check specific channel
wachi check -c "slack://xoxb-token/channel"

For detailed behavior (dedup model, error patterns, notification format, config schema), see references/spec.md.