llama-index-retrievers-dewey 0.1.0

LlamaIndex retriever integration for Dewey — managed RAG backend

LlamaIndex Dewey Retriever

Dewey is a managed document intelligence backend that handles the full RAG pipeline — PDF conversion, section extraction, chunking, embedding, and hybrid retrieval — behind a single REST API.

This package provides a DeweyRetriever that integrates Dewey's hybrid semantic + BM25 search directly into LlamaIndex pipelines.

Installation

pip install llama-index-retrievers-dewey

Usage

from llama_index.retrievers.dewey import DeweyRetriever
from llama_index.core.query_engine import RetrieverQueryEngine
from llama_index.llms.anthropic import Anthropic

retriever = DeweyRetriever(
    api_key="dwy_live_...",       # Dewey project API key
    collection_id="3f7a1b2c-...", # Collection to search
    k=8,                          # Number of chunks to retrieve
)

# Use with any LlamaIndex query engine
query_engine = RetrieverQueryEngine.from_args(
    retriever=retriever,
    llm=Anthropic(model="claude-haiku-4-5-20251001"),
)
response = query_engine.query("What are the main findings?")
print(response)

Direct retrieval

nodes = retriever.retrieve("attention mechanism scaling")
for n in nodes:
    print(f"[{n.score:.3f}] {n.node.metadata['filename']} › {n.node.metadata['section_title']}")
    print(f"  {n.node.text[:120]}...")

As part of an agent

from llama_index.core.tools import RetrieverTool
from llama_index.core.agent import ReActAgent
from llama_index.llms.openai import OpenAI

tool = RetrieverTool.from_defaults(
    retriever=retriever,
    description="Search AI research papers for relevant information.",
)
agent = ReActAgent.from_tools([tool], llm=OpenAI(model="gpt-4o-mini"), verbose=True)
agent.chat("Compare the Transformer and RAG architectures.")

Parameters

Parameter	Type	Default	Description
api_key	str	required	Dewey project API key (dwy_live_...)
collection_id	str	required	UUID of the collection to search
k	int	10	Max chunks to retrieve (1–50)
base_url	str	https://api.meetdewey.com/v1	Override for local dev

What Dewey returns

Each retrieved node includes:

• node.text — chunk content
• node.metadata["filename"] — source document filename
• node.metadata["section_title"] — section heading
• node.metadata["section_level"] — heading depth (1 = top-level)
• node.metadata["document_id"] — document UUID
• node.metadata["section_id"] — section UUID
• score — relevance score from hybrid RRF ranking

Resources

• Dewey documentation
• Free tier signup — no credit card required
• Python SDK (meetdewey)