Overview
Ansa helps documentation teams:
- Provide instant answers from your docs
- Reduce support tickets for documented topics
- Surface relevant articles naturally
- Identify documentation gaps from user questions
- Handle questions your docs don’t cover gracefully
Knowledge Base Setup
Indexing Your Documentation
Ansa supports multiple ways to index docs:
| Source Type | Best For |
|---|
| Web Crawl | Public documentation sites |
| File Upload | PDFs, markdown files, internal docs |
| Text Sources | Custom content, snippets |
| Q&A Pairs | Common questions with vetted answers |
Web Crawling Tips
When crawling documentation sites:
- Start from the docs root (e.g.,
docs.yoursite.com)
- Set appropriate depth — Most docs need 3-5 levels
- Exclude irrelevant paths — Changelog, release notes (unless needed)
- Re-sync regularly — Set up scheduled syncs for updated content
Use the auto-detect feature during onboarding. Ansa automatically discovers docs pages at /docs, /documentation, /help, /guides, and /api.
Structuring Content for Best Results
Good documentation indexing:
✅ Clear headings and subheadings
✅ Concise paragraphs
✅ Code examples with context
✅ Descriptive link text
Avoid:
❌ Single-page docs with everything
❌ Heavily nested accordions (may not crawl)
❌ Content loaded via JavaScript (use static/SSR)
Agent Configuration
System Prompt for Documentation
You are a helpful documentation assistant for [Product Name].
Your role:
- Answer questions accurately based on the documentation
- Provide code examples when relevant
- Link to relevant documentation pages for further reading
- Be concise but thorough
When you don't know the answer:
- Acknowledge honestly that you couldn't find this in the docs
- Suggest related topics that might help
- Offer to connect them with support if needed
Format responses for readability:
- Use markdown formatting
- Include code blocks with proper syntax highlighting
- Use bullet points for lists
Temperature Settings
| Use Case | Temperature | Why |
|---|
| API Reference | 0.1 - 0.3 | Accuracy is critical |
| Tutorials/Guides | 0.3 - 0.5 | Balance of accuracy and explanation |
| Conceptual Docs | 0.5 - 0.7 | More natural explanations |
Model Selection
For documentation assistants:
| Model | Best For |
|---|
| Claude 3.5 Sonnet | Excellent at explaining code and concepts |
| GPT-4o | Great for technical documentation |
| GPT-4o-mini | Cost-effective for high-volume sites |
Suggested Messages
Guide users to common topics:
Configure suggested messages for quick access:
{
"suggestedMessages": [
"How do I get started?",
"Show me authentication examples",
"What are the API rate limits?",
"How do I handle errors?"
]
}
Placement
For documentation sites:
| Placement | Position | When |
|---|
| Bottom Right | Default | General docs sites |
| Bottom Left | Alternative | If conflicts with ToC |
| Inline | Embedded | Specific help sections |
Triggers for Documentation
Engagement on Complex Pages
Help users on lengthy or complex docs:
{
"name": "Long Page Help",
"triggerType": "time_on_page",
"conditions": {
"timeOnPage": 60,
"page": "/docs/advanced/*"
},
"actionType": "show_bubble",
"actionConfig": {
"message": "Need help understanding this? I can explain any part.",
"duration": 8
}
}
Search Page Fallback
Engage users who might not find what they need:
{
"name": "Search Assistance",
"triggerType": "time_on_page",
"conditions": {
"timeOnPage": 30,
"page": "/docs/search*"
},
"actionType": "open_chat",
"actionConfig": {
"message": "Can't find what you're looking for? Tell me what you need and I'll help you find it."
}
}
Handling Unknown Questions
When the agent can’t find an answer in the docs:
1. Honest Acknowledgment
System prompt should include:
When information isn't in the documentation:
- Say "I couldn't find this in our documentation"
- Suggest similar topics that might help
- Offer to connect them with support
{
"name": "docs_feedback",
"description": "Use when you cannot answer the user's question from the documentation.",
"executionType": "form",
"formSchema": {
"fields": [
{
"name": "question",
"label": "Your Question",
"type": "textarea",
"defaultValue": "{{context.lastMessage}}"
},
{
"name": "email",
"label": "Email (for follow-up)",
"type": "email"
}
],
"submitLabel": "Submit Question",
"successMessage": "We'll add this to our docs or get back to you!"
},
"formPostActions": [
{
"type": "slack",
"webhookUrl": "https://hooks.slack.com/...",
"messageTemplate": "📝 *Docs Gap Identified*\n\n{{question}}\n\nFrom: {{email}}"
}
]
}
Identifying Documentation Gaps
Review Low-Confidence Conversations
Use the analytics dashboard to find:
- Questions with low confidence scores
- Repeated questions not well-answered
- Topics users ask about that aren’t documented
Create Q&A Pairs
For frequently asked questions not well-covered:
- Go to Knowledge Base → Q&A
- Add the exact question users ask
- Provide the ideal answer
- These get high priority in searches
GitBook, Docusaurus, Mintlify
Web crawl your public docs URL. Ensure:
- Content is server-rendered
- No robots.txt blocking
- Sitemap available for best coverage
Notion, Confluence
For internal docs:
- Export as PDF or markdown
- Upload via file sources
- Set up periodic re-uploads for updates
API Reference (OpenAPI)
Upload your OpenAPI/Swagger spec as a file source. The agent will understand:
- Endpoints and methods
- Request/response schemas
- Authentication requirements
See Also
