knowledge-base-builder_skill

Overview

This skill builds and optimizes elizaOS knowledge bases using document ingestion, embeddings, and semantic retrieval for RAG-enabled agents. It guides structure, chunking, embedding generation, vector storage, and search tuning to make agent knowledge production-ready. The goal is reliable, fast, and maintainable semantic search for agent conversations and tools.

How this skill works

It ingests documents (Markdown, PDF, text), applies chunking strategies (fixed, semantic, sliding window), and generates embeddings in batches. Chunks and embeddings are stored in a vector store and served via semantic search with hybrid ranking (embeddings + keyword). It includes tools for ongoing updates, versioning, and quality metrics to monitor relevance and latency.

When to use it

• Create a domain-specific knowledge base for an elizaOS agent
• Build a RAG system from a large set of documents or manuals
• Setup semantic search for agent responses and tool grounding
• Migrate or version knowledge while preserving retrieval performance

Best practices

• Use clear headers and consistent document structure for semantic chunking
• Choose chunk sizes between 500–1,500 characters and include 10–20% overlap
• Pre-compute embeddings and batch requests to reduce latency and cost
• Version knowledge files and record last-updated dates for traceability

Example use cases

• Ingest API docs and tutorials so the agent can answer developer questions with source citations
• Build a game design knowledge base for asset generation rules, workflows, and examples
• Implement semantic QA over technical manuals to reduce time-to-answer for support agents
• Keep training data current by scheduling embeddings regeneration when documents change