CodeCanyon Comment Assistant

Overview

CodeCanyon Comment Assistant is an automated AI pipeline built by Iqonic Design that monitors public CodeCanyon product pages for un-replied customer comments, generates context-aware draft replies using a Retrieval-Augmented Generation (RAG) pipeline powered by LlamaIndex and Gemini 2.0 Flash, and delivers the draft to the developer via a formatted Telegram message for final review before posting.

---

Architecture

Customer Comment on CodeCanyon
        ↓
Playwright Scraper (guest, no login)
        ↓
SQLite (duplicate check)
        ↓
LlamaIndex RAG (retrieves relevant context)
        ↓
Gemini 2.0 Flash (generates draft reply)
        ↓
Telegram Bot (notifies developer)
        ↓
Developer reviews → copies → pastes on CodeCanyon

---

Project Structure

codecanyon-comment-assistant/
├── main.py                          # Entry point — CLI run or Flask serve mode
├── requirements.txt                 # Python dependencies
├── ecosystem.config.js              # PM2 process config for production
├── .env                             # Secrets (git-ignored)
├── .env.example                     # Template — copy to .env and fill in values
├── .gitignore
├── scraper/
│   ├── codecanyon_scraper.py         # Playwright guest scraper — finds un-replied comments
│   └── comment_store.py              # SQLite deduplication store — tracks comment status
├── rag/
│   ├── doc_crawler.py                # One-time setup — crawls docs site to .txt files
│   ├── qa_extractor.py               # One-time setup — scrapes past Q&A reply pairs
│   ├── indexer.py                    # Builds and persists per-product LlamaIndex vector index
│   ├── retriever.py                  # Queries the index and returns top-k context chunks
│   └── knowledge_base/               # Per-product knowledge base files
│       ├── <product-id>/             # One folder per product (matches products.json id)
│       └── past_qa/                  # Shared past Q&A pairs (.txt)
├── ai/
│   └── draft_generator.py            # Gemini API call with system prompt — returns draft reply
├── notifier/
│   ├── telegram_notifier.py          # Sends HTML-formatted message via Telegram Bot API
│   └── mattermost_notifier.py        # Phase 2 alternative — Mattermost webhook
├── config/
│   └── products.json                 # List of CodeCanyon products to monitor
└── n8n/
    └── workflow_export.json          # Import into n8n to schedule the pipeline every 6 hours

---

Setup Instructions

1. Clone the repo

git clone <repo-url>
cd codecanyon-comment-assistant

2. Create virtual environment

python -m venv venv

3. Activate the virtual environment

# macOS / Linux
source venv/bin/activate

# Windows
venv\Scripts\activate

4. Install dependencies

pip install -r requirements.txt

5. Install Playwright browsers

playwright install chromium

6. Copy .env.example to .env and fill in all values

cp .env.example .env
# Edit .env and fill in GEMINI_API_KEY, TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID

7. Add product config to config/products.json

{
  "products": [
    {
      "id": "streamit-laravel",
      "name": "Streamit Laravel - OTT, Movies & Live Video Streaming Platform",
      "codecanyon_url": "https://codecanyon.net/item/streamit-laravel-movie-tv-show-video-streaming-platform-with-laravel/54895738/comments",
      "knowledge_base_path": "rag/knowledge_base/streamit-laravel/"
    }
  ]
}

Field reference:

Field	Description
id	Unique slug used for the index subfolder and logging
name	Display name shown in Telegram notifications
codecanyon_url	Full URL to the CodeCanyon item page (where comments appear)
knowledge_base_path	Path to the folder containing docs/changelogs for this product

8. Run the doc crawler for each product (one-time setup)

Crawls a product documentation site and saves the content as .txt files that the RAG indexer will ingest:

python rag/doc_crawler.py <base_url> <output_dir>

# Example:
python rag/doc_crawler.py https://docs.example.com rag/knowledge_base/streamit-flutter/

9. Run the Q&A extractor for each product (one-time setup)

Scrapes previously replied CodeCanyon comment threads and saves them as Q&A pairs for the RAG knowledge base:

python rag/qa_extractor.py <codecanyon_comments_url>

# Example:
python rag/qa_extractor.py https://codecanyon.net/item/streamit/12345678

10. Run the pipeline

python main.py run

11. Or start the Flask server for n8n

python main.py serve
# Starts on http://localhost:5001
# n8n calls GET http://localhost:5001/run

---

Environment Variables

Variable	Required	Description	Default
GEMINI_API_KEY	✅	Google Gemini API key	—
GEMINI_MODEL		Gemini generation model name	gemini-2.0-flash
GEMINI_EMBEDDING_MODEL		Gemini embedding model for RAG indexing	models/gemini-embedding-001
TELEGRAM_BOT_TOKEN	✅	Telegram bot token from @BotFather	—
TELEGRAM_CHAT_ID	✅	Telegram group/channel chat ID	—
AUTHOR_USERNAME		CodeCanyon seller username — used to detect own replies	iqonicdesign
DB_PATH		Path to SQLite database file	comment_store.db
INDEX_PERSIST_DIR		Directory where LlamaIndex vector indices are persisted	rag/index_store/

---

Phase 2 — Mattermost

The pipeline is designed to make swapping the notification channel trivial. notifier/mattermost_notifier.py is already built and ready. To switch from Telegram to Mattermost, add MATTERMOST_WEBHOOK_URL to your .env, then in main.py replace the send_to_telegram import and call with send_to_mattermost from notifier/mattermost_notifier.py — the function signature is identical. No other changes are needed.

---

Status Flow (SQLite)

pending  →  drafted  →  notified

Each comment_id is stored once; duplicate runs are safe.

---

Tuning the AI Tone

Open ai/draft_generator.py and edit the SYSTEM_PROMPT variable at the top of the file. This is intentionally separated from the generation logic so tone adjustments require no code changes.

---

n8n Setup Instructions

The workflow in n8n/workflow_export.json runs the pipeline every 6 hours automatically and sends a Telegram summary after each run.

Install n8n (self-hosted)

npm install -g n8n
n8n start

n8n runs on http://localhost:5678 by default.

Import the Workflow

1. Open n8n at http://localhost:5678
2. Go to Settings → Import Workflow
3. Select n8n/workflow_export.json
4. Click Import

Set Environment Variables in n8n

1. Go to Settings → Environment Variables
2. Add the following two variables:

Variable	Value
TELEGRAM_BOT_TOKEN	Your Telegram bot token
TELEGRAM_CHAT_ID	Your Telegram group/channel ID

Activate the Workflow

1. Open the imported workflow
2. Toggle the Active switch to ON
3. The pipeline will now run automatically every 6 hours (IST timezone)

Manual Test Run

1. Make sure the Flask server is running: python main.py serve
2. In n8n, open the workflow
3. Click Execute Workflow to trigger manually
4. Check Telegram for the summary message

---

Running in Production

Both the Flask server and n8n must be running persistently for automated scheduling to work. Use PM2 to keep both alive across reboots.

Install PM2

npm install -g pm2

Start both processes

pm2 start ecosystem.config.js

Start a single process

pm2 start ecosystem.config.js --only codecanyon-assistant
pm2 start ecosystem.config.js --only n8n

Save process list and enable auto-start on Windows boot

pm2 save
npm install -g pm2-windows-startup
pm2-startup install

Note: pm2 startup is Linux/macOS only. On Windows use pm2-windows-startup as shown above.

Useful PM2 commands

pm2 list                          # Show all processes and status
pm2 logs codecanyon-assistant     # Tail Flask server logs
pm2 logs n8n                      # Tail n8n logs
pm2 restart codecanyon-assistant  # Restart the Flask server
pm2 stop n8n                      # Stop n8n
pm2 delete all                    # Remove all processes from PM2