Bee Report Generation

Bee can automatically generate comprehensive, professionally formatted reports when users request in-depth analysis. Reports are processed asynchronously, allowing users to continue working while Bee researches and writes.

Overview

When you ask Bee for detailed analysis (typically 1,500+ words or ~2,000 tokens), the system automatically:

1. Detects the request - AI classifies your query as needing a full report
2. Queues generation - Creates a background job for the report worker
3. Researches the topic - Uses web search and Honey Jar context
4. Generates content - Produces comprehensive, well-structured analysis
5. Exports as PDF - Creates a beautifully formatted document
6. Notifies you - Report appears in your Reports dashboard

Classification Threshold

The system uses 2,000 tokens (~1,500 words) as the threshold:

• < 2,000 tokens → Handled as inline chat response
• ≥ 2,000 tokens → Queued as background report

When to Use Reports

✅ Good Candidates for Reports

• Comprehensive analysis - “Provide a detailed analysis of zero-trust architecture”
• Research documents - “Write a report on emerging cybersecurity threats”
• Compliance guides - “Create a comprehensive HIPAA compliance checklist”
• Technical deep-dives - “Explain microservices patterns with implementation examples”
• Business cases - “Develop a business case for adopting AI-powered security”

❌ Better as Chat Responses

• Quick questions - “What is STING?”
• Lists or summaries - “List the main features”
• How-to guides - “How do I create a Honey Jar?”
• Simple explanations - “Explain what PII means”

Triggering Report Generation

Automatic Detection

Bee classifies requests based on keywords that indicate expected length:

Explicit word counts also trigger reports:

• “Write a 2000 word analysis…”
• “Give me a 1500-word report…”

Example Prompts

These will trigger report generation:

"Provide a comprehensive analysis of SCADA/ICS security best practices, 
including NERC CIP compliance and network segmentation strategies."

"Write a detailed report on implementing zero-trust architecture 
in enterprise environments, with case studies and ROI analysis."

"Create a thorough technical breakdown of AI-powered threat detection 
systems with implementation examples."

Manual Trigger

You can also force report generation by saying:

• “Generate this as a report”
• “Create a PDF document for this”
• “Queue this as a long-form analysis”

Report Types

Bee automatically selects the appropriate report style based on your query:

Use Case Report

Triggered by: “how can”, “implement”, “deploy”, “use case”

Focuses on practical implementation scenarios with:

• Real-world applications
• Step-by-step implementation guides
• Industry-specific examples
• Best practices and pitfalls

Comparison Report

Triggered by: “compare”, “versus”, “vs”, “difference”

Side-by-side analysis including:

• Feature comparisons
• Pros and cons tables
• Decision frameworks
• Recommendations

Summary Report

Triggered by: “summarize”, “overview”, “brief”

Executive-level summaries with:

• Key findings
• High-level recommendations
• Quick reference sections
• Action items

Technical Report

Triggered by: “technical”, “architecture”, “deep dive”

Detailed technical analysis with:

• Architecture diagrams
• Code examples
• Implementation details
• Performance considerations

The Generation Process

┌─────────────────────────────────────────────────────────────┐
│                    User Request                              │
│  "Provide a detailed analysis on..."                        │
└─────────────────────┬───────────────────────────────────────┘
                      │
                      ▼
┌─────────────────────────────────────────────────────────────┐
│                 AI Classification                            │
│  • Analyzes query for report indicators                     │
│  • Estimates output length (tokens)                         │
│  • Determines report type (use_case, comparison, etc.)      │
└─────────────────────┬───────────────────────────────────────┘
                      │
          ┌───────────┴───────────┐
          │  < 2,000 tokens?      │
          └───────────┬───────────┘
                      │
         YES ←────────┴────────→ NO
          │                       │
          ▼                       ▼
   ┌──────────────┐    ┌──────────────────────┐
   │ Chat Response│    │ Queue Report Job     │
   │ (immediate)  │    │ (background worker)  │
   └──────────────┘    └──────────┬───────────┘
                                  │
                                  ▼
                       ┌──────────────────────┐
                       │ Report Worker        │
                       │ • Web search context │
                       │ • Honey Jar RAG      │
                       │ • LLM generation     │
                       │ • PDF export         │
                       └──────────────────────┘

Timeline

Generation time depends on:

• Report length requested
• Topic complexity
• Current queue depth
• LLM provider speed

Web Search Enhancement

When web search is enabled, Bee enriches reports with:

• Current information - Latest news, updates, and trends
• External sources - Citations and references
• Broader context - Industry perspectives beyond your knowledge base

See Web Search Integration for configuration details.

Honey Jar Integration

Reports can leverage your Honey Jars for domain-specific context:

1. Specify a Honey Jar when creating the report
2. RAG retrieval finds relevant documents
3. Context injection adds your knowledge to the prompt
4. Citation includes references to your documents

This creates reports that combine:

• 🌐 Web research (current external information)
• 📚 Your knowledge base (domain expertise)
• 🤖 AI analysis (synthesis and recommendations)

Viewing Reports

Reports Dashboard

Access your reports at /reports:

┌─────────────────────────────────────────────────────────────┐
│ 📄 Securing Power Grid SCADA/ICS with STING-CE             │
│                                                             │
│ Status: ✅ Completed                                        │
│ Generated: 12 minutes ago                                   │
│ Size: 2.4 MB (4,200 words)                                 │
│                                                             │
│ [📥 Download PDF]  [🗑️ Delete]                              │
└─────────────────────────────────────────────────────────────┘

Report States

Output Formats

Best Practices

Writing Effective Prompts

✅ Do:

• Use trigger words (“comprehensive”, “detailed”, “in-depth”)
• Include specific topics to cover
• Mention the intended audience
• Request specific sections or structure

❌ Don’t:

• Be vague (“tell me about security”)
• Request unrealistic lengths
• Expect real-time response for reports

Example: Well-Structured Request

"Generate a comprehensive report on implementing zero-trust 
architecture for a mid-sized financial services company.

Include:
1. Executive summary for C-suite
2. Technical architecture overview
3. Implementation phases and timeline
4. Vendor comparison (Microsoft, Zscaler, Cloudflare)
5. Compliance considerations (SOX, PCI-DSS)
6. ROI analysis and cost projections

Target audience: IT leadership and security teams."

PII Protection

Reports automatically:

• Detect PII in generated content
• Scramble sensitive data (configurable)
• Audit log PII handling decisions

Enable/disable via scrambling_enabled parameter.

Troubleshooting

Report Stuck in “Processing”

1. Check queue status at /reports
2. Reports typically complete in 4-20 minutes
3. Complex topics may take longer
4. Check system logs if exceeding 30 minutes

Report Failed

Common causes:

• LLM provider timeout
• Insufficient context for topic
• Rate limiting on web search

Solution: Click “Retry” to requeue the report.

Chat Response Instead of Report

If Bee responds inline instead of queuing a report:

• Add length indicators: “comprehensive”, “detailed”, “in-depth”
• Explicitly request: “Generate this as a report”
• Specify word count: “Write a 2000 word analysis…”

API Access

Reports can also be created programmatically. See the Reports API Reference for:

• Creating reports via API
• Polling for completion
• Downloading results

Configuration

Environment Variables