Ask

What is Ask

Ask is OctoReport's intelligent Q&A feature, allowing you to interact with the system in natural language to quickly explore knowledge base content or have general conversations.

Core Features:

• Two Modes: Library Mode + General Mode
• Streaming Responses: Real-time feedback, character-by-character display, no waiting
• Reasoning Process: Support Chain-of-Thought display
• Keyboard Shortcuts: Cmd/Ctrl+K to quickly start new conversation

PlaceholderAsk interface screenshot - showing two modes and Ask UI

---

Library Mode

What is Library Mode

Library Mode is based on RAG (Retrieval-Augmented Generation) technology, allowing AI to answer questions based on your knowledge base content, ensuring accurate and verifiable answers.

How it Works:

Your question
  ↓
Keyword extraction (LLM)
  ↓
Retrieve relevant content (Top 10)
  ↓
Build context
  ↓
Generate answer (with citations)
  ↓
Stream output

How to Use

Step 1: Select Library

1. Click "Library Mode" at the top
2. Select target library from dropdown (e.g., "AI Industry News")
3. Library must contain content to use

Step 2: Ask Questions

Enter your question, for example:

• "What important AI news has there been recently?"
• "What were the tech highlights this past week?"
• "Summarize OpenAI's latest developments"

Step 3: View Answer

AI will:

1. Retrieve relevant content from library (up to 10 items)
2. Generate answer based on content
3. Cite original sources (title + URL)
4. Stream output character by character (real-time feedback)

Tip: Library Mode answers are based on collected content. If library is empty or content irrelevant, AI cannot provide effective answers.

Retrieval Mechanism

Keyword Extraction:

• AI automatically extracts 3-5 keywords from your question
• Example: "Recent AI large model progress" →

  ["AI", "large model", "progress"]

Content Matching:

• Search library for content containing keywords
• Priority: Title match > Full-text match
• Return most relevant Top 10 items

Context Building:

[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

Generate Answer:

• LLM generates answer based on context
• Auto-cite sources (showing titles and links)

Best Practices

✅ Specific questions get better answers

Good examples:

• "What major AI models were released this week?"
• "What are the key features of GPT-4?"
• "Which companies are hiring AI engineers?"

Poor examples:

• "What's new?" (Too vague)
• "Tell me everything" (Too broad)

✅ Ensure library has relevant content

• Library Mode depends on collected content
• If empty or irrelevant, consider:
  1. Adding more data sources
  2. Adjusting collection frequency
  3. Switching to General Mode

✅ Use follow-up questions

Ask supports context memory:

• First question: "What important AI news is there today?"
• Follow-up: "Which company is most noteworthy?" (AI understands you're asking about the news from previous answer)

---

General Mode

What is General Mode

General Mode is universal conversation mode, not limited to knowledge base content, suitable for:

• General questions (e.g., "How to write Python code?")
• Brainstorming (e.g., "How to design a report template?")
• Explanations (e.g., "What is RAG?")

Difference from Library Mode:

• Library Mode: Answer based on your knowledge base
• General Mode: Answer based on LLM's built-in knowledge

How to Use

1. Click "General Mode" at top
2. Enter any question
3. AI will answer based on built-in knowledge

Use Cases:

• Technical consultation: "How to optimize LLM prompt?"
• Knowledge explanation: "What is the difference between GPT-4 and Claude?"
• Creative brainstorming: "Design a competitive intelligence report structure"

---

Advanced Features

Web Search (General Mode)

When enabling Web Search:

• AI will search web for latest info
• Cite search result sources
• Cost: 10 credits/search

How to Enable:

1. Switch to General Mode
2. Toggle "Enable Web Search" switch
3. Ask question (e.g., "What did OpenAI announce today?")

When to Use:

• Need latest information (recent days/today)
• Knowledge base has no relevant content
• Need to verify facts

Chain-of-Thought Display

Some models (like o1, DeepSeek) support displaying reasoning process:

• Show AI's thinking steps
• Help understand answer logic
• Useful for learning and debugging

---

Cost & Performance

Library Mode Cost

Per Conversation:

• Keyword extraction: ~1 credit
• Content retrieval: Free (database query)
• Answer generation: 5-20 credits (depends on model)

Total: ~5-20 credits/conversation

General Mode Cost

Per Conversation:

• Answer generation: 5-50 credits (depends on model and length)

With Web Search: +10 credits

Performance Optimization

Reduce Cost:

• Library Mode: Use fast models (e.g., GPT-4o-mini)
• Limit conversation length (shorter prompts)
• Disable Web Search when not needed

Improve Quality:

• Use premium models (e.g., GPT-4o, Claude Sonnet)
• Ensure library has enough relevant content
• Ask specific questions

---

Troubleshooting

Problem: Library Mode returns "No relevant content"

Possible Causes:

1. Library is empty
2. Keywords don't match content
3. Content all marked as expired

Solution:

• Check library content count
• Try different phrasing
• Check data source collection status

Problem: Answer quality insufficient

Possible Causes:

1. Retrieved content not relevant
2. Question too vague
3. Model choice inappropriate

Solution:

• Ask more specific questions
• Increase data source collection frequency
• Switch to premium model

Problem: Stream response interrupted

Possible Causes:

1. Network unstable
2. Backend service timeout
3. Browser compatibility issue

Solution:

• Refresh and retry
• Check network connection
• Contact admin to check backend service status

---

Next Steps

• Report Generation - Auto-generate structured reports on schedule
• Trigger Inbox - Trigger reports via natural language email
• Credits & Logs - View consumption details and task logs