𝕏

𝕏-Bot

An Autonomous AI-Powered Twitter/X Agent

"Intelligent content creation with personality-driven engagement"

Personal Project | 2025

The Problem

Why build an autonomous social media bot?

⏰ Time-Intensive

Consistent social media presence requires significant daily time investment

📊 Suboptimal Timing

Manual posting often misses optimal engagement windows

🌊 Content Overload

Engaging with relevant content is overwhelming at scale

🤖 AI Opportunity

Leverage LLMs for personality-consistent, high-quality content

The Solution

X Bot: Key Capabilities

Feature	Description
🤖 Autonomous Posting	Generates original tweets based on configurable personality
📖 Intelligent Reading	Scans timeline, identifies posts matching interests
💡 Inspiration Engine	Creates content inspired by curated interesting posts
🔔 Notification Handling	Monitors mentions and replies for engagement
🧠 Memory System	Avoids duplicate content using vector embeddings
📊 Web Dashboard	Real-time monitoring, analytics, settings, and live control
💬 Reply Workflows	Checks mentions/replies and queues notifications for processing

System Architecture

🕒

Scheduler

APScheduler

→

⚙️

Jobs Layer

Posting • Reading • Notifications • Replies • Inspiration

→

𝕏

X Automation

Selenium + undetected-chromedriver + stealth

🧩

LangChain + LangGraph

Generation, validation, reading, notifications workflows

↔

🤖

LLM Providers

OpenRouter • OpenAI • Google • Anthropic

↔

🧠

ChromaDB Memory

Embedding cache + duplicate detection (0.85)

💾

State Storage

JSON runtime state + SQLite analytics

→

🌐

FastAPI Dashboard

Jinja2 + HTMX + Chart.js + live settings reload

Personality System

Configurable bot personality via YAML

personality:
  tone: humorous
  style: conversational
  topics:
    - AI
    - technology
    - crypto
    - blockchain
    - web3
    - memes
  min_tweet_length: 60
  max_tweet_length: 280

Personality Enforces:

Brand alignment checking before posting
Tweet re-evaluation for quality control
Interest detection based on configured topics
Consistent voice across all generated content

Content Pipeline

From timeline reading to intelligent posting

📖 Read Timeline

Selenium extraction

→

🔍 Interest Check

LLM evaluation

→

⭐ Queue

Store interesting

→

💡 Generate

Inspiration batch

→

✅ Quality Check

Re-evaluation

→

📤 Post

Publish to X

                        Batch Processing: Interesting posts accumulate until threshold (25 posts), 
                        then the bot generates inspired content based on the collection.
                    

Memory & Deduplication

ChromaDB vector storage for intelligent memory

🧠 ChromaDB Features

Every tweet embedded before storage
Cosine similarity for duplicate detection
Configurable threshold: 0.85
Metadata persisted: timestamps, provider, operation
Backed by data/chroma/ for persistence

Duplicate Prevention: Semantically similar content is blocked before posting

# Duplicate detection logic
async def check_duplicate(text: str):
    embedding = await get_embedding(text)
    results = collection.query(
        query_embeddings=[embedding],
        n_results=1
    )
    
    if results["distances"][0][0] > 0.85:
        logger.warning("duplicate_detected")
        return True
    return False

Web Dashboard

Full monitoring and configuration interface

📊 Dashboard

Bot status, uptime, activity stats, rate limit progress

📝 Posts

Read, written, rejected, and interesting posts with tabs

💬 Chat

Interactive tweet generation with post-to-X option

📈 Analytics

Hourly token usage graph with Chart.js, provider breakdown

📋 Logs

Real-time log viewer with filtering and auto-scroll

⚙️ Settings

Live configuration editing with instant reload

                        Built with: FastAPI + Jinja2 + HTMX for dynamic partial loading • X.com-inspired dark/light themes
                    

Rate Limits & Compliance

Enforced Limits

Limit	Value
Max posts/day	`15`
Max replies/day	`50`
Reset time	`00:00 UTC`
Action delay	`1-3 seconds` (randomized)

Compliance Features

Rate limit enforcement per day
Human-like randomized delays
Cookie persistence (no repeated logins)
Bot disclosure in profile (🤖 AI Bot)
undetected-chromedriver to avoid detection
Stealth mode with User-Agent rotation

Current Status

Roadmap and repository snapshot

✅ Phase 1: Scheduler & Reading Complete

✅ Phase 2: Interest & Reactions Complete

🔶 Phase 3: Notifications & Reply Flow In Progress

✅ Phase 4: Memory & Dashboard Complete

4

LLM Providers

6

Scheduler Jobs

5

Graph Workflows

8

Dashboard Templates

Live Demo

Dashboard walkthrough

1️⃣ Dashboard Overview

Bot status, uptime, rate limits, activity feed

2️⃣ Posts Tab

Read/written/rejected/interested posts with pagination

3️⃣ Analytics

Token usage graph, provider breakdown, hourly stats

4️⃣ Settings

Live configuration with instant scheduler reload

🖥️ Open Dashboard: http://localhost:8000

Tech Stack

Layer	Technology
Language	`Python >= 3.13` with full type hints
AI Orchestration	LangChain + LangGraph
Providers	OpenRouter, OpenAI, Google, Anthropic
Configured Model	`x-ai/grok-4.1-fast` (OpenRouter)
Embeddings	`text-embedding-3-small`
Vector DB	ChromaDB (persistent local storage)
Browser	undetected-chromedriver + selenium-stealth
Scheduling	APScheduler (background scheduler)
State & Analytics	JSON state + SQLite (`aiosqlite`)

🌐 Web Layer

FastAPI backend
Jinja2 templates + HTMX partials
Chart.js token analytics
Pydantic v2 + pydantic-settings

                                Repo Facts: 6 scheduler jobs, 5 LangGraph workflows, and async I/O across state and database layers.
                            

Challenges & Learnings

Challenge	Solution
X/Twitter bot detection	`undetected-chromedriver` + selenium-stealth
LLM provider reliability	Multi-provider fallback (OpenRouter → OpenAI → Google)
Rate limiting enforcement	Per-day counters with UTC reset logic
Duplicate content prevention	Vector embeddings + cosine similarity threshold
Async complexity	Full `async/await` architecture with aiosqlite
Configuration management	YAML config + Pydantic validation + live reload

Future Roadmap

Planned enhancements

🔴 Intent Detection - Filter negative/toxic replies High Priority

🔴 Reply Generation - Respond to positive notifications High Priority

🟡 Trends Integration - Topic-aware content with compliance Medium

🟡 RSS Feed Ingestion - External content sources Medium

🟢 Multi-account Support - Manage multiple bots Low Priority

🟢 Cost Tracking - Per-provider API spend analytics Low Priority

Key Takeaways

🏗️ Modular Architecture

Clean separation of concerns with dedicated modules for each responsibility

🔒 Type Safety

Pydantic v2 models throughout for validation and documentation

🔄 Multi-Provider LLM

Resilient AI integration with automatic fallback between providers

👁️ Observable

Full dashboard with analytics, logs, and real-time status

✅ Compliant

Rate limits, bot disclosure, human-like behavior patterns

🔌 Extensible

Easy to add new features, providers, and integrations

❓

Questions?

Thank you for your attention!

📂 Source Code

Available on request

📧 Contact

Your contact info here