Skip to main content

⚡ Quick Start: Your Help Agent in 30 Minutes

🎯 Goal: Test Help Agent in Your Account Right Now​

Let's get you up and running with a private help agent using your existing infrastructure - no code changes needed initially!

✅ 30-Minute Workflow​

Step 1: Create Private Help Agent (5 minutes)​

Using Your Existing Agent UI​

1. Go to your agent creation interface
2. Create new agent with these settings:

Name: "Help Agent (Testing)"
Description: "Testing documentation-powered help for Bike4Mind features"

Trigger Words: @help, @docs, @assistant

Personality:
- Major Motivation: "Helping users understand Bike4Mind features"  
- Quirk: "I love connecting documentation to real user problems"
- Flaw: "Sometimes I provide too much detail when testing"

⚠️ IMPORTANT: Keep isPublic = false for now (private testing)

Tags: help-system, testing, documentation

Step 2: Upload Documentation (10 minutes)​

Prepare Documentation Files​

# 1. Zip your docs-site content
cd /path/to/docs-site/docs
zip -r bike4mind-docs.zip . -x "node_modules/*" "*.git*"

# 2. Upload via your existing Knowledge interface
# - Upload bike4mind-docs.zip
# - Add tags: "help-docs", "official-docs", "agent-knowledge"  
# - Let your existing chunking/vectorization process it

Key Files to Include​

✅ Features documentation (mementos.md, quest-master.md, etc.)
✅ Getting started guides
✅ Technical documentation
✅ Troubleshooting guides
✅ Voice agents documentation (your latest docs!)
✅ Security and admin guides

Step 3: Test Basic Functionality (10 minutes)​

Test Queries in Your Chat​

# Basic feature questions
"@help what are Mementos"
"@docs how do I use Questmaster"
"@assistant explain voice agents"

# Troubleshooting queries  
"@help voice agents not working"
"@docs troubleshoot file upload"

# Getting started queries
"@help getting started guide"
"@docs how to share files"
"@assistant best practices"

Check Agent Response Quality​

✅ Does agent respond to @help triggers?
✅ Does response include relevant documentation?
✅ Does agent maintain helpful personality?
✅ Are sources cited when available?
✅ Is response quality good enough for public use?

Step 4: Quick Iteration (5 minutes)​

Tune Based on Results​

# If responses are too long:
- Adjust personality: "I provide concise, actionable answers"

# If missing documentation:
- Check file tags: ensure "help-docs" tag is applied
- Check chunking: verify files were processed

# If agent doesn't respond:
- Check trigger words: ensure @help, @docs, @assistant
- Check agent status: ensure agent is active

🔧 Zero-Code Enhancement (Optional)​

If You Want Better Document Search Right Away​

Add This Simple Enhancement​

// You can add this to your existing AgentDetectionFeature.ts
// This works for YOUR private agent immediately

async getContextMessages(quest, embeddingFactory, message, maxTokens, modelInfo) {
  const baseContext = await super.getContextMessages(...arguments);
  
  // Check if this is a help agent (yours!)
  const agentsToProcess = (quest as any)._agentsToProcess || [];
  const isHelpAgent = agentsToProcess.some(agent => 
    agent.triggerWords?.some(trigger => 
      ['@help', '@docs', '@assistant'].includes(trigger)
    )
  );
  
  if (isHelpAgent) {
    // Search YOUR help docs using existing infrastructure
    const helpContext = await this.searchYourHelpDocs(message, embeddingFactory);
    return [...baseContext, ...helpContext];
  }
  
  return baseContext;
}

async searchYourHelpDocs(message, embeddingFactory) {
  const queryEmbedding = await embeddingFactory.createEmbedding(message);
  
  // Search YOUR files tagged as help-docs
  const relevantChunks = await fabFileChunkRepository.findByVectorSimilarity({
    vector: queryEmbedding,
    limit: 5,
    threshold: 0.7,
    filters: {
      'fabFile.userId': this.service.user.id, // YOUR files
      'fabFile.tags': { $in: ['help-docs', 'official-docs'] }
    }
  });
  
  if (relevantChunks.length === 0) {
    return [{ role: 'system', content: 'No specific documentation found. Provide general help.' }];
  }
  
  const context = ['Available documentation:'];
  relevantChunks.forEach(chunk => {
    context.push(`Source: ${chunk.fabFile?.filename}`);
    context.push(`Content: ${chunk.text}`);
    context.push('---');
  });
  
  return [{ role: 'system', content: context.join('\n') }];
}

🎉 When Ready to Go Public​

Flip the Switch (2 minutes)​

1. Edit your tested agent
2. Change isPublic to true
3. Change name to "Bike" (or whatever you prefer)
4. Update description: "I'm Bike4Mind's official assistant..."
5. Add tags: "official", "public", "system-agent"
6. Save changes

🎉 Now EVERYONE can use "@help" or "hey bike" to get documentation help!

📊 Success Indicators​

Immediate (First Hour)​

  • Agent responds to @help triggers in your chat
  • Responses include content from uploaded documentation
  • Agent personality comes through in responses
  • No error messages or system issues

Short Term (First Day)​

  • Response quality is good enough to share publicly
  • Agent handles various types of questions well
  • Documentation coverage is comprehensive
  • Ready to flip to public

Medium Term (First Week)​

  • Other users discover and use the public agent
  • Positive feedback on response quality
  • Reduced simple questions in support channels
  • Agent becomes part of user workflow

🚨 Troubleshooting​

"Agent doesn't respond to @help"​

✓ Check trigger words include "@help"
✓ Verify agent is active/enabled
✓ Test with simple message: "@help test"
✓ Check server logs for agent detection

"No documentation in responses"​

✓ Verify files uploaded with "help-docs" tag
✓ Check chunking completed (wait ~5-10 minutes after upload)
✓ Test vector search working: search for known content
✓ Verify AgentDetectionFeature enhancement if added

"Response quality is poor"​

✓ Adjust agent personality to be more concise
✓ Upload more specific/targeted documentation
✓ Test with different types of questions
✓ Consider adjusting vector search threshold

🚀 Next Steps​

After Successful Testing​

  1. Go Public: Flip isPublic: true when ready
  2. Announce: Let your team know about "@help" command
  3. Monitor: Track usage and gather feedback
  4. Iterate: Improve based on user interactions
  5. Scale: Consider the marketplace vision for community agents

Future Enhancements​

  • Natural language detection ("Hey Bike")
  • Voice integration ("Hey Bike, how do I...")
  • Proactive help suggestions
  • Usage analytics and optimization
  • Community agent sharing

💡 Why This Works​

✅ Uses 100% existing infrastructure - no new systems needed ✅ Risk-free testing - private agent in your account first
✅ Immediate value - working help agent in 30 minutes ✅ Easy to scale - flip to public when ready ✅ Foundation for more - sets up agent marketplace future

Ready to start? Create your help agent now! 🚀