Skip to main content

FAQ

This document addresses common questions about knowledge sources in Kapa.

Does adding an LLMs.txt file improve accuracy?​

Kapa does not currently support the LLMs.txt standard. The platform already includes built-in features that address similar goals:

  • Project-specific context configuration
  • Retrieval mechanisms for finding relevant sources
  • Optimized parsing of documentation content

Kapa's existing functionality provides solutions for accuracy without requiring this file. The platform may consider supporting LLMs.txt if it becomes more widely adopted in the industry.

How should I format markdown files for AI ingestion?​

For optimal AI comprehension, markdown files must be properly formatted with clear structure and appropriate use of markdown elements. Poorly formatted documents significantly degrade answer quality.

Required formatting​

  • Use proper heading hierarchy (# Title, ## Section, ### Subsection)
  • Include meaningful headings that describe the content
  • Use standard markdown syntax for lists, code blocks, and emphasis
  • Avoid documents that are purely code blocks without explanatory text

Common formatting issues to avoid​

  • Documents without any headings or structure
  • Files containing only code blocks with no context or explanation
  • Pseudo-structured formats (like key-value pairs) instead of proper markdown
  • Content where meaning depends on visual formatting rather than semantic structure

Example: Support ticket format​

Instead of unstructured key-value pairs:

Title: Connection timeout issue
Description: Customer experiencing timeouts when connecting to API
Tags: api, timeout, connectivity
CreatedBy: user@example.com

Use proper markdown structure:

# Support Ticket: Connection timeout issue

**Timestamp**: 2025-01-15 14:30 UTC  

**Status**: Resolved  

**Tags**: api, timeout, connectivity

## Issue description

Customer experiencing timeouts when connecting to API endpoint.

## Conversation

**User**: I'm getting timeout errors when trying to connect to the API.

**Agent**: I can help you troubleshoot the connection timeout. Let me check...

For detailed guidance on writing content optimized for AI systems, see our comprehensive Writing documentation for AI: best practices guide.

Can I connect another MCP server to Kapa as a knowledge source?​

No. Kapa exposes your knowledge sources as an MCP server. It does not act as an MCP client that consumes other MCP servers as a data source.

To add knowledge to a Kapa project, use one of the native data source connectors. See Data sources overview for the full list.

Can I connect Kapa to my SQL database or other structured data?​

Connecting directly to backend databases is not supported out of the box. Database schemas and data structures vary widely across organizations, making a generic integration impractical.

If you are trying to access this kind of data in the context of building an agent for your customers, the Agent SDK is the right approach. The Agent SDK supports custom tools that execute client-side, allowing your agent to query your database or any other data source using your own logic and authentication context. This gives you full control over what data the agent can access and how it is retrieved.

What languages can I index and search in?​

You can index knowledge in arbitrary languages and search it in arbitrary languages, and the two do not have to match: a user can ask a question in French and Kapa can answer it, in French, from English knowledge sources. This multilingual capability comes from the underlying language models, which are multilingual by design. Kapa is optimized for conversations in English but handles a wide variety of languages well.

To change the language of the Website Widget's UI itself, see the widget's language configuration.

Can I upload sources or trigger a refresh of sources using the API?​

No, we do not provide an API endpoint for uploading or managing sources. Kapa automatically does data source refreshes. To add, delete, or manually refresh sources, use the Sources tab on the Kapa platform.

You can always reach out to the Kapa team for help.