Skip to content

algorithm-builder.md

Latest commit

 

History

History
280 lines (211 loc) · 9.43 KB

algorithm-builder.md

File metadata and controls

280 lines (211 loc) · 9.43 KB
Raw

Algorithm Builder — 24/7 Niche Growth Automation

Train your X algorithm around specific topics by mimicking natural human behavior — with optional LLM-powered comments.

Overview

The Algorithm Builder is a long-running browser automation that systematically interacts with content in your niche to shape your X algorithm. It runs continuously with work/break cycles, engaging with posts the way a real person would:

  • Search your niche keywords (Top + Latest tabs)
  • Scroll feeds naturally with variable speed and reading pauses
  • Like relevant posts (probability-based)
  • Comment with LLM-generated or curated replies
  • Follow viral accounts that match your niche
  • Browse home feed, Explore, target accounts, and your own profile
  • Rest between sessions (20-45 min work, 5-20 min break)

Two ways to run it:

  1. Standalone script — paste core.js + algorithmBuilder.js into DevTools
  2. Control Panel — select "🧠 Algorithm Builder" from the floating panel UI

Quick Start (Standalone)

  1. Open x.com → DevTools Console
  2. Paste src/automation/core.js
  3. Edit the config sections in src/automation/algorithmBuilder.js:
    • Set your KEYWORDS, TARGET_ACCOUNTS, BIO_KEYWORDS
    • (Optional) Set LLM_CONFIG.ENABLED = true and add your API_KEY
  4. Paste the script
  5. Let it run — it handles breaks automatically
// Stop anytime
window.stopAlgoBuilder()

// Check stats
window.algoStats()

// View config
window.algoConfig()

// Reset all state
window.algoReset()

Quick Start (Control Panel)

  1. Paste core.js, then controlPanel.js
  2. Select 🧠 Algorithm Builder from the dropdown
  3. Fill in: keywords, persona, bio filters, target accounts
  4. (Optional) Paste your OpenRouter API key for LLM comments
  5. Click ▶ Start

Configuration

NICHE_CONFIG

const NICHE_CONFIG = {
  // Your identity (used in LLM system prompt)
  PERSONA: 'a crypto & web3 builder who shares alpha and builds in public',

  // Keywords the builder searches for
  KEYWORDS: ['web3 builder', 'crypto alpha', 'DeFi yield', 'solana ecosystem'],

  // Accounts whose posts you want to engage with
  TARGET_ACCOUNTS: ['@nichxbt', '@VitalikButerin'],

  // Only follow users whose bio contains at least one of these
  BIO_KEYWORDS: ['crypto', 'web3', 'defi', 'nft', 'blockchain', 'builder'],

  // Topics to look for on Explore page
  EXPLORE_TOPICS: ['Technology', 'Crypto', 'Business'],
};

LLM_CONFIG

const LLM_CONFIG = {
  ENABLED: false,                    // Set to true to enable AI comments
  API_URL: 'https://openrouter.ai/api/v1/chat/completions',
  API_KEY: '',                       // Your OpenRouter or OpenAI API key
  MODEL: 'google/gemini-flash-1.5',  // Any OpenRouter model
  SYSTEM_PROMPT: '...',              // Customizable persona prompt
  MAX_TOKENS: 80,
  TEMPERATURE: 0.9,                  // Higher = more creative
  LLM_PROBABILITY: 0.7,             // 70% chance to use LLM vs fallback
};

Any OpenAI-compatible endpoint works — just change API_URL.

BEHAVIOR

const BEHAVIOR = {
  // Session timing
  SESSION_MIN_MINUTES: 20,
  SESSION_MAX_MINUTES: 45,
  BREAK_MIN_MINUTES: 5,
  BREAK_MAX_MINUTES: 20,

  // Peak hours (UTC) — higher activity during these hours
  PEAK_HOURS_START: 13,   // 1 PM UTC
  PEAK_HOURS_END: 23,     // 11 PM UTC

  // Action probabilities per tweet (0-1)
  LIKE_PROBABILITY: 0.35,
  COMMENT_PROBABILITY: 0.08,
  RETWEET_PROBABILITY: 0.05,
  FOLLOW_PROBABILITY: 0.15,

  // Off-peak multiplier (reduces all probabilities)
  OFF_PEAK_MULTIPLIER: 0.4,

  // Per-session limits
  MAX_LIKES_PER_SESSION: 25,
  MAX_COMMENTS_PER_SESSION: 5,
  MAX_FOLLOWS_PER_SESSION: 15,
  MAX_RETWEETS_PER_SESSION: 5,

  // Per-day hard limits
  MAX_LIKES_PER_DAY: 150,
  MAX_COMMENTS_PER_DAY: 25,
  MAX_FOLLOWS_PER_DAY: 80,
  MAX_RETWEETS_PER_DAY: 30,

  // Follow filters
  MIN_FOLLOWERS_TO_FOLLOW: 10,
  MAX_FOLLOWERS_TO_FOLLOW: 50000,
  SKIP_PROTECTED_ACCOUNTS: true,
};

Activity Cycles

The builder randomly selects activities using weighted probabilities:

Cycle Weight What Happens
🔍 Search & Engage 30% Search a keyword → scroll Top + Latest → like, comment, follow
🏠 Browse Home 25% Scroll the "For you" feed, engage naturally
🎯 Target Account 15% Visit a configured target's profile, engage with their posts
🌍 Explore 10% Browse trending topics, click into trends
👥 Search People 15% Search users by keyword, follow those matching bio filters
👤 Own Profile 5% Visit your own profile briefly (looks natural)

This rotation mimics how real users browse X — they don't just like 500 posts in a row.

LLM Comment Generation

When enabled, comments are generated by an LLM that reads the tweet and replies in context:

System prompt structure:

You are {PERSONA} on Twitter/X.
Generate a short, authentic reply to the tweet below.
Rules:
- 1-2 sentences max, casual Twitter tone
- No hashtags, no emojis unless very natural
- Be specific to the tweet's content, not generic
- Sound like a real person, not a bot
- Occasionally ask a genuine question
- Never start with "Great post!" or similar generic openers

Flow:

  1. Tweet text extracted from the page
  2. 70% chance to send to LLM (configurable)
  3. If LLM fails or skipped → use curated fallback comment
  4. Comment typed character-by-character with random delays

Fallback Comments

When LLM is disabled or fails, one of 20 curated comments is used:

"This is a really solid take"
"Been thinking about this a lot lately"
"Underrated perspective"
"Interesting — what made you think of this?"
"This tracks with what I've been researching"

These are intentionally varied in tone and length to avoid detection.

Human Behavior Patterns

The builder employs multiple strategies to appear natural:

Pattern Implementation
Variable delays All timings randomized within ranges (2-6s between actions)
Reading pauses 15% chance of 5-12 second "reading" pause while scrolling
Peak/off-peak Action probabilities drop 60% during off-peak hours
Session breaks 5-20 min breaks between 20-45 min sessions
Scroll variation Scroll distance randomized 300-800px per step
Typing jitter Comments typed at ~60ms/char with ±20ms random variance
Mixed activities Weighted random cycle selection — not repetitive
Profile visits Occasionally visits own profile (5% weight)

State Persistence

All state is saved to localStorage and survives page refreshes:

  • Total likes, comments, follows, retweets across all sessions
  • Last 500 engaged tweet IDs (deduplication)
  • Last 500 followed usernames (deduplication)
  • Last 500 commented tweet IDs (no double-commenting)
  • Keyword and target account rotation indices
  • Session start time and break history

Run window.algoReset() to clear all state and start fresh.

Console API

Command Description
window.stopAlgoBuilder() Stop after the current action finishes
window.algoStats() Print all-time stats and return stats object
window.algoReset() Clear all saved state
window.algoConfig() Print current configuration
window.XActions.AlgoBuilder Full namespace with state, config, and methods

Stats Output

╔═══════════════════════════════════════════════════╗
║       📊 Algorithm Builder — All Time Stats       ║
╠═══════════════════════════════════════════════════╣
║  Runtime:    4.2 hours                             ║
║  Sessions:   8                                     ║
║  Searches:   24                                    ║
║  ❤️  Likes:    187                                  ║
║  💬 Comments: 14                                    ║
║  ➕ Follows:  62                                    ║
║  🔁 Retweets: 11                                   ║
║  Engaged:    214 unique tweets                     ║
╚═══════════════════════════════════════════════════╝

Rate Limits

The builder uses XActions' built-in rate limiting (rateLimit.check() / rateLimit.increment()) to enforce both per-session and per-day caps. When daily limits are hit, those action types are skipped — the builder continues other activities.

Action Per Session Per Day
Likes 25 150
Comments 5 25
Follows 15 80
Retweets 5 30

Tips

  • Start small: On a fresh account, lower all limits by 50% for the first week
  • Niche down: 3-5 focused keywords work better than 20 broad ones
  • Use LLM sparingly: COMMENT_PROBABILITY: 0.08 means ~8% of tweets get a comment — this is intentional
  • Target accounts matter: Pick 3-5 thought leaders in your niche whose audiences you want to attract
  • Monitor your account: Watch for rate limit warnings in the X UI — if you see them, increase delays
  • Don't run multiple scripts: One automation at a time to avoid rate limit stacking