langchain-keiro

Official LangChain integration for the Keiro deep search API with embeddings.

All components use deep mode with embeddings enabled — no configuration needed. Every request returns full content plus 768-dimensional embedding vectors (configurable to 384/512/1024 via Matryoshka).

Installation

pip install langchain-keiro

Quick Start

Retriever

Use Keiro as a LangChain retriever for RAG pipelines:

from langchain_keiro.retrievers import KeiroRetriever

retriever = KeiroRetriever(
    keiro_api_key="keiro_...",
    max_results=3,
)

# Sync
docs = retriever.invoke("What is artificial intelligence?")

# Async
docs = await retriever.ainvoke("What is artificial intelligence?")

# Each doc includes embeddings in metadata:
# doc.metadata["query_embedding"]  — 768-dim vector for the query
# doc.metadata["chunk_embeddings"]  — per-chunk embeddings with text
# doc.metadata["tldr"], ["tags"], ["entities"], etc.

# Use in a RAG chain
from langchain_openai import ChatOpenAI
from langchain.chains import create_retrieval_chain

llm = ChatOpenAI(model="gpt-4")
chain = create_retrieval_chain(retriever, llm)
response = chain.invoke({"input": "What is AI?"})

Embeddings

Generate embedding vectors directly:

from langchain_keiro.embeddings import KeiroEmbeddings

embeddings = KeiroEmbeddings(
    keiro_api_key="keiro_...",
    dimensions=768,  # Matryoshka: 384, 512, 768, or 1024
)

# Single query
vec = embeddings.embed_query("What is AI?")
# vec is a list of 768 floats

# Batch documents
vecs = embeddings.embed_documents(["doc 1", "doc 2", "doc 3"])

# Async
vec = await embeddings.aembed_query("What is AI?")
vecs = await embeddings.aembed_documents(["doc 1", "doc 2"])

Tool (for Agents)

Use Keiro as a tool in LangChain agents:

from langchain_keiro.tools import KeiroSearchTool

tool = KeiroSearchTool(keiro_api_key="keiro_...")

# Direct invocation
result = tool.invoke({"query": "latest AI research"})

# With an agent
from langchain_openai import ChatOpenAI
from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain_core.prompts import ChatPromptTemplate

llm = ChatOpenAI(model="gpt-4")
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a research assistant. Use the keiro_search tool to find information."),
    ("human", "{input}"),
    ("placeholder", "{agent_scratchpad}"),
])
agent = create_tool_calling_agent(llm, [tool], prompt)
agent_executor = AgentExecutor(agent=agent, tools=[tool])
response = agent_executor.invoke({"input": "What are the latest AI trends?"})

Configuration

KeiroRetriever

Always uses mode="deep" with embeddings.enabled=true.

KeiroEmbeddings

KeiroSearchTool

Always uses mode="deep" with embeddings.enabled=true.

Embedding Dimensions (Matryoshka)

The Keiro API supports Matryoshka embedding dimensions, allowing you to trade off vector size for precision:

• 384: Smallest, fastest, least precise
• 512: Good balance for lightweight use
• 768 (default): Standard precision
• 1024: Maximum precision

Response Enrichment

The deep mode response includes rich metadata beyond just content:

API Key

Get your API key at keirolabs.cloud. The Explorer plan includes 500 free credits.

Development

# Install for development
pip install -e ".[test]"

# Run unit tests
pytest tests/unit_tests/

# Run integration tests (requires KEIRO_API_KEY)
KEIRO_API_KEY=keiro_... pytest tests/integration_tests/

License

MIT