Report Generation

What is Report Generation

Report generation is one of OctoReport's core features, automatically creating structured analysis reports from knowledge base content using AI.

Core Features:

• Multi-step Processing: Break complex reports into multiple steps, each focused on one task
• Model Combination: Different steps can use different LLM models (balance cost vs quality)
• Variable Injection: Dynamically fetch latest data, auto-fill into prompts
• Scheduled Generation: Set weekly plans for auto-generating daily/weekly/monthly reports
• Version Management: Keep all historical reports, view past versions anytime

PlaceholderReport generation workflow - Library → Multi-step processing → Final report

Report Template Concept

What is a Report Template

A report template defines how to generate reports, including:

1. Basic Info

   • Template name (e.g., "Daily AI Industry News Summary")
   • Associated libraries (which libraries to extract data from)
   • Schedule strategy (when to auto-generate)

2. Report Steps (ReportStep)

   • Each template contains 1-5 steps
   • Steps execute in order, later steps can use earlier outputs
   • Each step can select different LLM models

3. Prompt Templates

   • Support variable injection (

     {{variable}}

     )
   • Support JSON format data injection
   • Flexible output format control

Template vs Report Relationship

Report Template (ReportTemplate)
  ├─ Step 1: Data aggregation (GPT-4o-mini, quick summary)
  ├─ Step 2: Deep analysis (GPT-4o, high-quality analysis)
  └─ Step 3: Format output (GPT-4o-mini, Markdown format)
       ↓
Generated Report (Report)
  ├─ Step 1 result: News list summary
  ├─ Step 2 result: Industry trend analysis
  └─ Step 3 result: Complete Markdown report

ℹ️ Tip: One template can generate multiple reports, each generation is fully recorded.

Multi-Step LLM Processing

Why Multi-Step

Single-step processing problems:

• ❌ Prompt too long, model easily "loses focus"
• ❌ Can't balance cost vs quality (either all expensive or all cheap models)
• ❌ Hard to debug (can't pinpoint which part fails)

Multi-step processing advantages:

• ✅ Each step focuses on one task, improves accuracy
• ✅ Flexible model combination (cheap for summary, expensive for analysis)
• ✅ Can debug each step individually
• ✅ Steps can pass data between each other

Typical Three-Step Design

Step 1: Data Aggregation (Fast model, e.g., GPT-4o-mini)

• Extract latest content from library
• Format as structured list
• Cost: ~50-100 credits

Step 2: Analysis & Insights (Premium model, e.g., GPT-4o/Claude)

• Analyze trends, extract insights
• Generate professional analysis
• Cost: ~200-500 credits

Step 3: Format Output (Fast model, e.g., GPT-4o-mini)

• Format as Markdown/HTML
• Add charts, tables
• Cost: ~50-100 credits

Total cost: ~300-700 credits, vs single-step premium model ~500-1000 credits

Variable System

Built-in Variables

OctoReport provides the following built-in variables, dynamically fetched when generating reports:

{{libraryContent}}

Description: All content from associated libraries (JSON format)

Format:

[object Object],
  ,[object Object],
    ,[object Object],[object Object], ,[object Object],[object Object],
    ,[object Object],[object Object], ,[object Object],[object Object],
    ,[object Object],[object Object], ,[object Object],[object Object],
    ,[object Object],[object Object], ,[object Object],[object Object],
    ,[object Object],[object Object], ,[object Object],[object Object],
    ,[object Object],[object Object], ,[object Object],[object Object],
    ,[object Object],[object Object], ,[object Object],
  ,[object Object],
,[object Object],
hljs json

Use case: Used in Step 1 for data aggregation

{{previousStepOutput}}

Description: Output from previous step (available from Step 2 onwards)

Use case: Pass Step 1 summary to Step 2 for analysis

{{currentDate}}

Description: Current date (YYYY-MM-DD format)

Example:

2025-01-20

{{currentTime}}

Description: Current time (HH:mm:ss format)

Example:

09:30:00

Custom Variables

You can define custom variables in template configuration:

Scenario: Weekly report needs to specify report period

Configuration:

[object Object],
  ,[object Object],[object Object], ,[object Object],
    ,[object Object],[object Object], ,[object Object],[object Object],
    ,[object Object],[object Object], ,[object Object],[object Object],
    ,[object Object],[object Object], ,[object Object],
  ,[object Object],
,[object Object],
hljs json

Usage in Prompt:

Generate {{reportType}} for {{reportPeriod}}, target audience: {{targetAudience}}

Schedule Strategy

Interval Mode

Execute every X hours:

• Example: Every 24 hours (daily report)
• Example: Every 168 hours (weekly report)

Configuration:

[object Object],
  ,[object Object],[object Object], ,[object Object],[object Object],
  ,[object Object],[object Object], ,[object Object],
,[object Object],
hljs json

Weekly Plan Mode

Specific day + time:

• Example: Mon/Wed/Fri 9:00
• Example: Every Monday 18:00

Configuration:

[object Object],
  ,[object Object],[object Object], ,[object Object],[object Object],
  ,[object Object],[object Object], ,[object Object],
    ,[object Object],[object Object], ,[object Object],[object Object],[object Object], ,[object Object],[object Object], ,[object Object],[object Object],[object Object],  ,[object Object],
    ,[object Object],[object Object], ,[object Object],
  ,[object Object],
,[object Object],
hljs json

Manual Trigger

No auto-execution, click "Generate Now" to trigger manually.

Best Practices

✅ Daily News Summary

Scenario: Aggregate 10-20 tech news daily, generate morning briefing

Template Design:

• Step 1: Extract titles, summaries (GPT-4o-mini)
• Step 2: Categorize news by topic (GPT-4o-mini)
• Step 3: Format as Markdown (GPT-4o-mini)

Schedule: Every day 7:00 AM

Cost: ~150 credits/report

✅ Weekly Analysis Report

Scenario: Analyze industry trends, generate investor insights

Template Design:

• Step 1: Aggregate weekly data (GPT-4o-mini)
• Step 2: Deep trend analysis (Claude Sonnet)
• Step 3: Generate investment advice (GPT-4o)

Schedule: Every Monday 9:00 AM

Cost: ~600 credits/report

✅ Real-time Monitoring Alert

Scenario: Monitor competitor moves, alert immediately

Template Design:

• Step 1: Extract key events (GPT-4o-mini)
• Step 2: Assess impact level (GPT-4o)

Schedule: Every 4 hours

Cost: ~200 credits/report

Troubleshooting

Problem: Generated report is empty

Possible causes:

1. Associated library has no content
2. Time range filter too strict
3. Content marked as expired

Solution:

• Check library content count
• Adjust time range (

  days

  parameter)
• Check dedup strategy

Problem: High credit consumption

Possible causes:

1. All steps use premium models
2. Input content too much
3. Prompt too long

Solution:

• Use fast models for aggregation steps
• Enable Smart Content Filter
• Simplify prompts, remove unnecessary context

Problem: Report quality insufficient

Possible causes:

1. Prompt unclear
2. Wrong model choice
3. Insufficient input data

Solution:

• Add specific output format requirements
• Use premium models for analysis steps
• Increase data source collection frequency

Next Steps

• Ask - Use Ask to quickly explore library
• Configuration Tips - Report template design best practices
• Credits & Logs - View detailed credit consumption