knowledge-base-ingestion

Knowledge Base Ingestion Skill

You have the ability to add content to the vector knowledge base. This allows you to build a searchable repository of information that can be retrieved later using semantic search.

Capabilities

1. Text Ingestion: Add plain text content directly to the knowledge base. The text is automatically chunked and embedded for semantic search.

2. URL Ingestion: Fetch content from a URL and add it to the knowledge base. Useful for adding web pages, documentation, or online resources.

3. Document Management: Delete documents that are no longer needed, or get information about existing documents.

When to Use This Skill

Use knowledge base ingestion when:

• The user wants to add information for later retrieval
• Building a custom knowledge repository
• Adding reference materials, documentation, or notes
• The user shares content they want to "remember" or store

Ingestion Guidelines

1. Use Descriptive Titles: Choose titles that will help identify the content later. Good titles make it easier to find documents.

2. Organize with Collections: Group related documents into collections for better organization:

   • technical_docs - Technical documentation
   • meeting_notes - Meeting summaries
   • research - Research materials
   • policies - Company policies

3. Chunk Size: Documents are automatically split into smaller chunks for better search relevance. You don't need to worry about document size.

4. Confirm Success: After ingestion, confirm the document was added successfully by reporting the document ID and chunk count.

Example Workflows

Adding Text Content

1. User provides text to remember
2. Call kb_ingest_text with the text and a descriptive title
3. Optionally specify a collection
4. Confirm success with document details

Adding Web Content

1. User provides a URL
2. Call kb_ingest_url with the URL
3. Content is fetched, processed, and stored
4. Confirm success with document details

Removing Content

1. User requests document removal
2. Use kb_list_documents (from KB Search skill) to find the document
3. Call kb_delete_document with the document ID
4. Confirm deletion

Important Notes

• Ingested content is stored in a vector database for semantic search
• Content is automatically chunked into smaller pieces for better retrieval
• Embeddings are generated using OpenAI's text-embedding model
• Deleted documents cannot be recovered