Use codemap find "SymbolName" to locate any symbol before reading — it gives exact file + line range and is faster than grep across the whole repo.

RAG Pipeline (Document Chunking + Embedding + Retrieval)

Overview

Documents
    │
    ▼  TextLoader / MarkdownLoader
    │
    ▼  FixedSizeChunker / SentenceChunker
    │
    ▼  InMemoryVectorStore (TF-IDF by default)
    │
    ▼  KnowledgeBase.search(query)
    │
    ▼  Agent tool call → grounded answer

Quick start

from lauren_ai._knowledge import KnowledgeBase, TextLoader, FixedSizeChunker
from lauren_ai._memory._vector import InMemoryVectorStore

kb = KnowledgeBase(
    store=InMemoryVectorStore(),
    chunker=FixedSizeChunker(chunk_size=512, overlap=64),
)

# Index documents
await kb.load(TextLoader("docs/faq.txt"))
await kb.load(TextLoader("The quick brown fox jumps over the lazy dog.", is_file=False))

# Search
results = await kb.search("quick fox", top_k=3)
for r in results:
    print(f"{r.score:.3f} — {r.content}")

Attaching KB to an agent (as a tool)

from lauren_ai import agent, use_tools

@agent(model="claude-opus-4-6", system="Answer questions using the knowledge base.")
@use_tools(kb.as_tool(name="search_company_docs", top_k=5))
class RAGAgent: ...

The as_tool() method returns a @tool()-decorated async function whose schema the LLM can call. Results include content, score, and any metadata stored at index time.

Attaching KB via KnowledgeSource (module-level)

from lauren_ai import agent, use_knowledge_sources, AgentModule
from lauren_ai._knowledge import KnowledgeBase, KnowledgeSource, TextLoader
from lauren_ai._memory._vector import InMemoryVectorStore

kb = KnowledgeBase(store=InMemoryVectorStore())
ks = KnowledgeSource(
    kb=kb,
    tool_name="search_docs",
    top_k=5,
    loaders=[TextLoader("docs/faq.txt")],  # loaded at app startup
)

@agent(model="claude-opus-4-6")
@use_knowledge_sources("search_docs")
class SupportAgent: ...

AIModule = AgentModule.for_root(
    agents=[SupportAgent],
    knowledge=[ks],
    imports=[LLMModule.for_root(cfg)],
)

Chunkers

Testing a RAG pipeline

Use InMemoryVectorStore directly in tests — no external dependencies needed. The TF-IDF implementation handles basic keyword matching out of the box.

kb = KnowledgeBase(store=InMemoryVectorStore())
await kb.load(TextLoader("Python is a programming language.", is_file=False))
results = await kb.search("programming language")
assert len(results) > 0
assert "Python" in results[0].content

For testing the full agent flow, queue a search tool call in MockTransport:

mock.queue_tool_use("search_company_docs", {"query": "refund policy"})
mock.queue_response(completion("Our refund policy allows 30-day returns."))

Reference

• lauren_ai._knowledge: KnowledgeBase, KnowledgeSource, TextLoader, FixedSizeChunker, SentenceChunker
• lauren_ai._memory._vector: InMemoryVectorStore
• lauren_ai._memory: MemoryResult
• Skills: managing-memory, vector-store-integration, embedding-model-ingestion