ChatLab uses Retrieval-Augmented Generation (RAG) to deliver answers based on your knowledge base. Understanding its limitations helps you optimize your chatbot's performance and set realistic expectations.
Context Window Limitations
ChatLab uses AI models with strict token limits for processing queries. The context window must fit the user question, conversation history, and retrieved knowledge combined.
Context sizes by plan:
- Free/Basic: 8,000 tokens
- Standard: 16,000 tokens
- Premium: 32,000 tokens
What happens when context is exceeded:
- Retrieved knowledge gets truncated
- Critical information may be omitted
- Answers may be incomplete or miss important details
Solutions:
- Remove redundant content from training data
- Split large files into smaller, topic-focused documents
- Upgrade your plan for larger context windows
- Use clear, specific role instructions to guide responses
Training and Content Extraction Limits
File Upload Limits
- Maximum file size: 50MB per file
- Supported formats: PDF, DOC, DOCX, TXT, CSV, XLS, XLSX
Content That Cannot Be Extracted
ChatLab cannot read or extract text from:
- Images and graphics - Text embedded in images is not recognized
- Scanned PDFs - PDFs containing images of text rather than actual text
- Login-protected pages - Pages requiring authentication
- Heavily JavaScript-rendered content - Some dynamic pages may not fully render
- Non-text media - Audio, video, or embedded files
Website Scanning Constraints
- Processing time: Large scans can take 10-60 minutes
- Deep nesting: Very deeply nested URLs may timeout
- Dynamic content: Some JavaScript frameworks prevent full content extraction
AI Behavior Challenges
Hallucination
AI models may generate plausible but incorrect responses when relevant information is not found in the knowledge base. The AI can draw from its general training data instead of admitting it does not know.
Reduce hallucination:
- Set Creativity to 0 (Bot Settings > Conversation > AI Behavior)
- Add explicit instructions in Role & Behavior settings: "Only respond based on the provided knowledge base. If you cannot find relevant information, say 'I don't have information about that.'"
- Keep your knowledge base comprehensive for your specific domain
Knowledge Base Awareness
ChatLab cannot:
- Summarize what is in the knowledge base
- List topics, files, or countries covered
- Report how many documents are trained
- Explain why specific information was retrieved
The chatbot retrieves relevant chunks for queries but does not understand the structure or metadata of your knowledge base.
Performance with Large Knowledge Bases
Semantic Overlap Issues
When multiple documents contain similar information, the system may:
- Retrieve less relevant content
- Miss the most appropriate answer
- Combine information from different contexts incorrectly
Solutions:
- Avoid duplicating content across files
- Use clear, distinct naming and structure
- Test queries that might match multiple sources
- Split comprehensive guides into focused topic files
Retrieval Prioritization
The AI retrieves content based on semantic similarity to the user's question. It does not:
- Prefer newer content over older
- Prioritize specific file types
- Understand document hierarchy or importance
Organize your knowledge base with distinct, non-overlapping content for best results.
Rate Limits and Quotas
Message Rate Limits
To prevent abuse, chatbots have configurable rate limits. When exceeded:
- Users see a rate limit message
- They must wait before sending more messages
- Configure limits in Bot Settings > Security
Monthly Message Credits
Each plan includes monthly message credits that reset on your billing date:
- Free: 30 messages
- Basic: 2,400 messages
- Standard: 11,000 messages
- Premium: 50,000 messages
When credits are exhausted, the chatbot stops responding until credits reset or you enable overage billing.
External Service Dependencies
ChatLab relies on third-party services:
- OpenAI/Google - AI model APIs for response generation
- Pinecone - Vector database for knowledge retrieval
- ZenRows - Website content extraction
Service disruptions or rate limits from these providers can temporarily affect:
- Response generation speed
- Website scanning capability
- Knowledge retrieval accuracy
These outages are rare but may cause brief delays or errors during maintenance windows.
Best Practices Summary
| Challenge | Solution |
|---|---|
| Incomplete answers | Reduce redundancy, upgrade context size |
| Hallucination | Set creativity to 0, add strict role instructions |
| Cannot extract content | Use text-based PDFs, avoid image-heavy sources |
| Overlapping results | Organize content into distinct topics |
| Rate limits hit | Adjust limits in Security settings |
| Slow training | Use sitemap scanning, reduce crawl depth |
For specific issues not covered here, contact support through Main Menu > Support Request.