Skip to content

Knowledge Base

OLAV includes a built-in semantic search knowledge base — you can import your team's operations documentation, runbooks, incident response playbooks, and more into OLAV, then search them using natural language.

Feature Claims

ID Claim Status
C-L2-21 Knowledge base indexes documents and supports semantic search (vector + BM25 hybrid) ✅ v0.10.0

Use Cases

  • "What is the standard procedure for BGP failover?"
  • "Where are the disaster recovery steps?"
  • "What was the solution used last time for this issue?"

No need to dig through Confluence, search the Wiki, or hunt through emails — just ask OLAV.

Importing Documents

Supported formats: .md (Markdown), .pdf, .txt. Place files in the .olav/knowledge/ directory, then index them:

olav --agent config "索引所有知识库文件"

Check indexing status:

olav --agent config "显示知识库状态"

Searching the Knowledge Base

Ask questions directly in natural language:

olav "我们的 Runbook 里关于 BGP 故障切换是怎么说的？"
olav --agent config "在知识库中搜索 'DR 流程'"

OLAV returns the most relevant document snippets and generates a comprehensive answer using the LLM.

How It Works

The knowledge base uses vector semantic search combined with BM25 keyword search (RRF fusion ranking). This approach understands semantic similarity while ensuring keyword matches are not missed:

Document → Chunking (~1024 chars) → Generate vector embeddings → Store in LanceDB
Query → Vector similarity search + BM25 keyword search → Fusion ranking → Return Top-N snippets → LLM synthesized answer

Configuring Vector Embeddings

Configure the embedding model in .olav/config/api.json:

"embedding": {
  "mode": "api",
  "api": {
    "model": "openai/text-embedding-3-small",
    "base_url": "https://openrouter.ai/api/v1"
  }
}


"embedding": {
  "mode": "local"
}
Uses the BAAI/bge model. CPU-only, no network connection required. Suitable for security-isolated environments.

Knowledge Base vs Agent Memory

The knowledge base stores documents you explicitly import. Agent memory (LanceDB) stores lessons learned automatically during OLAV operation (via /trace-review). Both support semantic search, but serve different purposes.