LangChain Using Tilores

This repository provides the building blocks for integrating LangChain, LangGraph, and the Tilores entity resolution system.

Developers can use these tools to create powerful systems that leverage entity resolution for record retrieval, search, and entity resolution.

Examples

• Human-in-the-Loop Chat: examples/chat

  This example demonstrates how to build a chat application using Chainlit and LangGraph to explore a Tilores instance through natural language. It guides users through search functionality and explains the search results.

• Basic Usage: examples/basic

  This example shows how to use tools with an LLM model in a basic setup.

Usage

from tilores import TiloresAPI
from langchain_tilores import TiloresTools
from langchain_openai import ChatOpenAI
from langchain_core.messages import HumanMessage

# Initialize the Tilores API.
tilores = TiloresAPI.from_environ()
# TiloresTools helps you build typed tools from a specific Tilores instance, typed according to
# the schema of the instance.
tilores_tools = TiloresTools(tilores)

# Setup a LLM model for inference bound with a set of tools.
tools = [tilores_tools.search_tool]
tools_dict = {tool.name: tool for tool in tools}
model = ChatOpenAI(temperature=0, streaming=True, model_name="gpt-4o")
model = model.bind_tools(tools)

# The basic loop works like this, that a list of messages is passed to the LLM
messages = [
    HumanMessage("Find me an entity by the first name Emma, surname Schulz, born on 1988-03-12")
]
ai_message = model.invoke(messages)
messages.append(ai_message)

# And for each AiMessage, you must check if it wants to invoke tools.
for tool_call in ai_message.tool_calls:
    # Perform the tool call and append the ToolMessage to the list of messages
    selected_tool = tools_dict[tool_call['name']]
    tool_message = selected_tool.invoke(tool_call)
    messages.append(tool_message)

# Then continue the basic loop by invoking the LLM with the current state, passing the list of messages.
ai_response = model.invoke(messages)
print(ai_response.content)

$ cd examples/basic/
$ pip install -r requirements.txt
$ python llm_with_tools.py
I found multiple records for an entity with the first name Emma, surname Schulz, born on 1988-03-12. Here are the details:

1. **Record ID:** cc001001-0006-4000-c000-000000000006
   - **First Name:** Emma
   - **Last Name:** Schulz
   - **Date of Birth:** 1988-03-12

2. **Record ID:** cc001001-0002-4000-c000-000000000002
   - **First Name:** Emma
   - **Last Name:** Schulz
   - **Date of Birth:** 1988-03-12

[... snip ...]

If you need more specific information or further assistance, please let me know!

Provided tools

• tilores_search