AI Reply DM

Automatically replies to unread Direct Messages using AI-generated contextual responses.

What It Does

Checks if the current inbox tab has unread messages (via tab indicator)
Finds the first unread conversation (scrolls down if needed)
Opens the conversation
Extracts the other user’s username from the thread header
Parses the last N messages to build conversation context
Sends conversation history to an AI provider for contextual reply generation
Types the reply with human-like simulation
Sends the reply
Verifies delivery (input emptied + send button disappeared)
Optionally returns to the inbox
Logs the action for tracking

Supported AI Providers

Provider	Models Available
OpenAI	GPT-4o, GPT-4o Mini, GPT-4 Turbo, GPT-4 Vision Preview
Claude (Anthropic)	Claude 3.5 Sonnet, Claude 3 Opus, Claude 3 Sonnet
DeepSeek	DeepSeek Chat, DeepSeek Reasoner
Gemini (Google)	Gemini 1.5 Pro, Gemini 1.5 Flash, Gemini Pro Vision
Groq	Llama 3.1 70B, Mixtral 8x7B
Custom	Any OpenAI-compatible endpoint

Configuration Options

Option	Type	Description
Account	Select	Account for logging
Provider	Select	AI provider to use
API Key	Password	Your API key (stored securely in node)
Model	Select	AI model (options depend on provider)
Custom Model Name	Text	For custom model names (when “Custom” selected)
Custom Endpoint URL	Text	For custom providers (OpenAI-compatible format)
Prompt	Textarea	Instructions for AI reply generation (up to 8000 chars)
Context Messages	Number	Number of recent messages to include as context (default: 5)
Max Scrolls to Find Unread	Number	Maximum scrolls to search for unread conversations (default: 5)
Min Length	Number	Minimum reply length in words (default: 1)
Max Length	Number	Maximum reply length in words (default: 500)
Blacklist	Textarea	Words to avoid, one per line
Enable Typing Simulation	Checkbox	Type like a human (recommended)
Typing Preset	Select	Fast, Normal, Slow, Human, or Custom
Custom Typing Config	Object	Min/max delay, chars per word (for Custom preset)
Go Back After Sending	Checkbox	Return to inbox after sending (default: on)
Wait After	Number	Seconds to wait after sending (default: 1.0)

Typing Presets

Preset	Speed	Realism
Fast	Quick	Low
Normal	Moderate	Medium
Slow	Slow	High
Human	Variable	Very High
Custom	Configurable	Configurable

How Unread Detection Works

The module detects unread messages at two levels:

Tab level: Checks if the inbox tab (Primary/General) has unread messages
Conversation level: Finds the unread dot indicator next to conversations, scrolling down the inbox if needed

How Message Parsing Works

Messages in a DM thread are differentiated by the presence of a sender avatar:

Has sender avatar — message received (from them)
No sender avatar — message sent (from you)

Only text messages are parsed. Non-text content (images, reels, voice notes) is skipped.

Features

Feature	Description
Conversation Context	AI receives the last N messages with [You]/[Them] labels for contextual replies
Unread-Only	Only processes conversations with unread messages
Smart Scrolling	Scrolls inbox to find unread conversations not visible on screen
Text-Only Generation	No screenshot needed — conversation parsed via XML for speed
Multi-Provider Support	6 AI providers + custom endpoints
Human-Like Typing	Configurable typing simulation with natural delays
Auto-Retry	Automatic retry on API failures (3 attempts, exponential backoff)
Blacklist Filtering	Avoid specific words in generated replies
Length Validation	Ensures replies meet min/max word requirements
Send Verification	Confirms reply was sent (input emptied + send button gone)
Go Back	Optionally returns to inbox after sending
Action Logging	Logs all replies to JSONL file for tracking

Data Storage

Data	Location
Reply Log	`Accounts/{account}/dms/dm_log.jsonl`

Log Entry Format

{
  "type": "reply_dm",
  "target_username": "jane_doe",
  "dm_text": "Generated reply text",
  "provider": "openai",
  "model": "gpt-4o-mini",
  "success": true,
  "account": "your_account",
  "timestamp": "2024-01-15T10:30:00Z"
}

Output Connections

This module has three outputs:

success — Reply was sent successfully
failed — An error occurred during execution
no_unread — No unread messages found in the current tab

When successful, the output includes:

provider: AI provider used
model: Model used
reply_text: The generated reply text
target_username: User replied to
messages_context: Number of messages used as context
duration: Execution time in seconds

Requirements

API key for chosen provider
Internet connection
Must already be in the DM inbox (use Navigate to Direct module before this)
At least one unread conversation

When to Use

Automatically responding to incoming DMs
Maintaining conversations at scale
Responding to customer inquiries or fan messages
Building engagement through timely replies

Example Prompt

You are replying to a DM on Instagram.
Keep the reply short, natural and conversational (1-3 sentences).
Match the tone of the conversation. Do not use hashtags.
If they asked a question, answer it helpfully.
Return ONLY the reply text.

Tips

Use the Navigate to Direct module before this one to ensure you’re in the inbox
Increase “Context Messages” for longer conversations that need more context
Customize your prompt to match your brand voice
Use blacklist to prevent unwanted words in replies
Enable “Go Back After Sending” to chain with Loop modules for multiple replies
Monitor logs to review AI-generated replies
Start with GPT-4o Mini or DeepSeek Chat for cost efficiency
Pair with a Loop module to reply to multiple unread conversations in sequence