Skip to main content

🧠 Smart Triggers

Smart Triggers filter user messages based on custom prompts to an AI (LLM). This powerful feature allows you to create sophisticated content moderation rules using natural language, leveraging the intelligence of various AI models to evaluate messages automatically.

Overview​

Unlike traditional triggers that rely on keyword matching, Smart Triggers use Large Language Models (LLMs) to understand context, intent, and meaning in messages. You define a yes/no question, and the AI evaluates each message to determine if it matches your criteria, then takes the actions you've configured.

Key capabilities:

  • Context-aware content filtering
  • Natural language rule definition
  • Support for multiple AI models (OpenAI, Google, Meta, Alibaba, and more)
  • Flexible action system
  • Optional targeting of new users only

Configuration​

To access Smart Triggers:

  1. Use the /settings command in your chat
  2. Navigate to 🤖 AI → 🧠 Smart triggers
  3. Toggle the feature 🟢 ON or 🔴 OFF

Main Settings​

Enabled/Disabled Toggle​

Enable or disable smart triggers for this community.

  • 🟢 Enabled: All configured smart triggers are active
  • 🔴 Disabled: All smart triggers are paused
Testing Phase

Keep the feature disabled while you're setting up and testing your triggers to avoid unintended actions.

AI Model​

Choose the AI model used to evaluate the prompt.

Fren One supports multiple AI models from different providers:

OpenAI Models​

  • GPT-4 (OpenAI): Most advanced OpenAI model, best accuracy and context understanding
  • GPT-3.5 Turbo (OpenAI): Fast and cost-effective, good for most use cases

Requirements: OpenAI API key (see API Keys section)

ASI:Cloud Models​

  • ASI Mini (ASI): Lightweight, fast inference
  • Gemma-3-27B (Google): Google's efficient LLM
  • Qwen3-32B (Alibaba): Alibaba's powerful language model
  • GPT OSS 20B (OpenAI): Open-source OpenAI model
  • Llama 3.3 70B Instruct (Meta): Meta's large instruction-tuned model
  • GLM 4.5 Air (Z.ai): Z.ai's optimized model
  • Mistral Nemo (Mistral): Mistral's efficient model

Requirements: ASI:Cloud API key (see API Keys section)

Model Selection
  • GPT-4: Best for complex reasoning, nuanced understanding, but higher cost
  • GPT-3.5 Turbo: Balanced performance and cost
  • ASI:Cloud models: Cost-effective alternative with various model sizes and capabilities

Triggers​

Manage your triggers for this community.

Features:

  • View all configured smart triggers
  • Create new triggers
  • Edit existing triggers
  • Delete triggers you no longer need
  • See trigger configuration at a glance

API Keys​

Smart Triggers require API keys from your chosen AI provider. You can add these through Chat Settings → API Keys.

OpenAI API Key​

Required for: GPT-4, GPT-3.5 Turbo, and other OpenAI models

How to get your key:

  1. Visit OpenAI Platform
  2. Sign up or log in
  3. Add billing details
  4. Navigate to API Keys section
  5. Create new API key
  6. Copy and add it to Fren One via Chat Settings → API Keys → OpenAI
API Key Security

Keep your API keys secure. Never share them publicly or commit them to version control.

ASI:Cloud API Key​

Required for: Gemma, Qwen, Llama, GLM, Mistral, GPT-OSS and other ASI:Cloud models

What is ASI:Cloud? ASI:Cloud (asicloud.cudos.org) is an affordable serverless inference platform that provides access to various open-source AI models.

How to get your key:

  1. Visit ASI:Cloud Documentation
  2. Follow the quickstart guide
  3. Sign up for ASI:Cloud using an email, Google account or permissionlessly with wallet
  4. Top up your balance using cryptocurrency payments
  5. Generate your API key
  6. Add it to Fren One via Chat Settings → API Keys → ASI:Cloud
Cost-Effective Alternative

ASI:Cloud offers more affordable inference compared to OpenAI, making it ideal for high-volume communities or budget-conscious projects.

Creating a Smart Trigger​

To create a new smart trigger:

  1. Go to Triggers menu
  2. Select + Create Trigger: Create a new AI-based trigger for this community
  3. Configure the following settings:

Trigger Configuration​

Prompt​

The yes/no question sent to the AI to decide whether to trigger an action.

Format: Frame your moderation rule as a question the AI can answer with yes or no.

Example prompts:

Does this message make reference to price action?
Is this message asking about when tokens will be listed on exchanges?
Does this message contain financial advice or investment recommendations?
Is this user promoting another cryptocurrency project?
Does this message include personal attacks or harassment?

Prompt writing tips:

  • Be specific and clear
  • Frame as a yes/no question
  • Focus on one concept per trigger
  • Test with various message types
  • Consider edge cases
Effective Prompts

Good prompts are specific, unambiguous, and focus on detectable characteristics. Instead of "Is this spam?", use "Does this message contain promotional links to external services?"

Actions​

The action to be taken when this trigger is triggered (when AI answers "yes").

Available actions:

  • Delete message: Remove the triggering message
  • Mute: Silence the user (set duration)
  • Kick: Remove user from the group (can rejoin)
  • Ban: Permanently ban user from the group
  • Warn: Issue a warning to the user
  • Reply: Send a reply to the triggering message
  • Send message: Send a standalone message to the chat

Multiple actions: You can combine actions for comprehensive responses.

Example combinations:

  • Warn + Send message: "Please refrain from discussing price here, thanks!"
  • Delete message + Mute: Remove and timeout
  • Warn + Reply: Direct feedback to user
  • Delete message + Send message: Remove and explain why
Action Severity

Start with less severe actions (warn, send message) and escalate only if needed. Test thoroughly before using ban or kick.

New users only​

Apply this trigger only to new users to reduce API usage.

Options:

  • ✅ Yes: Only evaluate messages from new users
  • ❌ No: Evaluate all non-admin messages

When to enable:

  • When the issue primarily affects new members (e.g., spam bots)
  • To reduce API costs for high-volume groups
  • When established members understand the rules

When to disable:

  • When the rule applies to everyone equally
  • For critical moderation rules (harassment, illegal content)
  • In smaller communities where API cost isn't a concern
Cost Optimization

Enabling "New users only" can significantly reduce API usage and costs, especially in active communities. Define "new user" criteria in your Filter settings.

Saving Your Trigger​

After configuring all settings:

  1. Click 💾 Save to activate the trigger
  2. The bot will confirm creation
  3. The trigger will appear in your triggers list

Managing Smart Triggers​

Viewing Triggers​

In the Triggers menu, you can see:

  • List of all configured smart triggers
  • Prompt for each trigger
  • Actions configured
  • New users only status

Example trigger display:

🧠 Smart trigger 1
💭 Prompt: Does this message discuss price action?
💥 Actions:
• warn
• Send message: Please refrain from discussing price here, thanks!
👶 New users only: ❌ No

Editing Triggers​

To modify an existing trigger:

  1. Select the trigger from the list
  2. Choose ✏️ Edit
  3. Modify prompt, actions, or settings
  4. Click 💾 Save

Deleting Triggers​

To remove a trigger:

  1. Select the trigger from the list
  2. Choose 🗑️ Delete
  3. Confirm deletion

Common Use Cases​

Preventing Price Discussion​

Scenario: Stop price speculation in main chat.

Configuration:

  • Prompt: "Does this message make reference to price action, price predictions, or ask about token price?"
  • Actions: Delete message + Send message
  • Message: "💰 Price discussion is not allowed here. Join #price-talk instead!"
  • New users only: No

Blocking Financial Advice​

Scenario: Prevent users from giving investment advice.

Configuration:

  • Prompt: "Does this message contain financial advice, investment recommendations, or tell people to buy/sell?"
  • Actions: Delete message + Warn
  • New users only: No

Redirecting Support Questions​

Scenario: Keep main chat clean by redirecting support to proper channel.

Configuration:

  • Prompt: "Is this message asking for technical support or help with an issue?"
  • Actions: Reply
  • Reply message: "👋 For technical support, please visit #support channel or open a ticket!"
  • New users only: Yes

Preventing Competitor Promotion​

Scenario: Stop promotion of competing projects.

Configuration:

  • Prompt: "Is this message promoting or recommending another cryptocurrency project or token?"
  • Actions: Delete message + Mute (1 hour)
  • New users only: No

Scenario: Prevent referral and affiliate link spam.

Configuration:

  • Prompt: "Does this message contain a referral link, affiliate link, or ask people to sign up using a code?"
  • Actions: Delete message + Warn
  • New users only: Yes

Detecting Misinformation​

Scenario: Flag potential misinformation for review.

Configuration:

  • Prompt: "Does this message make false claims about the project, team, or partnerships?"
  • Actions: Delete message + Send message (to mods channel)
  • Message: "⚠️ Potential misinformation detected - please review"
  • New users only: No

Best Practices​

Effective Smart Trigger Usage
  • Start Simple: Begin with clear, straightforward prompts
  • Test Thoroughly: Try various message types before enabling
  • Use Specific Language: Avoid ambiguous prompts
  • Monitor Performance: Track false positives and negatives
  • Combine with Traditional Triggers: Use both for comprehensive coverage
  • Document Your Prompts: Keep records of what each trigger does
  • Regular Review: Audit triggers monthly and adjust as needed
  • Consider Context: Account for legitimate uses of flagged terms
Avoiding Issues
  • Don't Over-Automate: Some situations need human judgment
  • Test Edge Cases: Consider how prompts handle borderline content
  • Monitor API Costs: Track usage, especially with GPT-4
  • Provide Explanations: Always tell users why content was removed
  • Human Oversight: Review AI decisions regularly
  • Privacy Awareness: Only process what's necessary for moderation
  • False Positives: Have a process to appeal AI decisions

Advanced Strategies​

Layered Moderation​

Combine different trigger types:

Layer 1 - Traditional Triggers:

  • Exact keyword matches
  • Known spam patterns
  • Blocked links

Layer 2 - Smart Triggers:

  • Context-aware content filtering
  • Intent detection
  • Nuanced rule enforcement

Layer 3 - Human Review:

  • Complex cases
  • Appeals
  • Policy updates

Prompt Engineering​

Write better prompts for accuracy:

Be Specific: ❌ "Is this spam?" ✅ "Does this message contain unsolicited promotional content for external products or services?"

Add Context: ❌ "Is this offensive?" ✅ "Does this message contain personal attacks, insults, or derogatory language directed at other members?"

Include Examples:

Does this message discuss price action? 
Examples of price action include:
- Asking "wen moon"
- Predicting future prices
- Discussing charts or technical analysis
- Asking when price will go up/down

Handle Edge Cases:

Does this message promote another project? 
Note: Mentioning other projects in context of comparison or discussion is OK, but actively encouraging people to invest in or join other projects is not OK.

Model Selection Strategy​

Choose models based on use case:

GPT-4 (High Accuracy):

  • Nuanced content requiring deep understanding
  • Complex policy enforcement
  • Multi-lingual content
  • Edge case handling

GPT-3.5 Turbo (Balanced):

  • Standard moderation tasks
  • High-volume communities
  • Cost-sensitive deployments
  • Quick response needed

ASI:Cloud Models (Cost-Effective):

  • Budget-conscious projects
  • Experimental triggers
  • High-volume testing
  • Specific model strengths (Llama for instruction-following, Gemma for efficiency)

Integration with Other Features​

Smart Triggers + Notifications:

  • Get notified when triggers activate
  • Track patterns in flagged content
  • Monitor AI accuracy

Smart Triggers + Warns:

  • First trigger → Auto-warn
  • Repeat triggers → Escalate action
  • Track violation history

Smart Triggers + User Whitelist:

  • Exempt trusted users from triggers
  • Reduce false positives
  • Allow moderators to discuss flagged topics

Troubleshooting​

Trigger not activating?

  • Verify Smart Triggers are enabled (🟢)
  • Check API key is configured correctly
  • Test the prompt with obvious examples
  • Ensure bot has necessary permissions for actions

Too many false positives?

  • Refine your prompt to be more specific
  • Add context and examples to prompt
  • Try a more advanced model (GPT-4)
  • Test prompt variations

Too many false negatives?

  • Broaden your prompt language
  • Include examples of what should be caught
  • Consider multiple triggers for related issues
  • Try a different AI model

High API costs?

  • Enable "New users only" where appropriate
  • Use ASI:Cloud instead of OpenAI
  • Use GPT-3.5 instead of GPT-4
  • Reduce number of active triggers
  • Combine similar prompts into one

Slow response times?

  • Choose faster models (GPT-3.5, ASI Mini)
  • Reduce complexity of prompts
  • Check API provider status
  • Consider caching for repeated content

API key not working?

  • Verify key is correctly copied (no extra spaces)
  • Check API credits/billing status
  • Ensure key has necessary permissions
  • Try regenerating the key
  • For ASI:Cloud, verify account is active

Messages in wrong language?

  • Some models handle specific languages better
  • Try different models for non-English content
  • Include language specification in prompt
  • Consider using multilingual models
tip

Smart Triggers are powerful but require proper setup and monitoring. Start with a few well-tested triggers and expand gradually. Always provide clear communication to your community about moderation policies.