langchain-prefid

Official LangChain integration for PrefID - Build preference-aware AI agents that adapt to user preferences

What is PrefID?

PrefID is identity-aware AI memory infrastructure that helps AI agents understand:

• WHAT users like (food, music, travel preferences)
• HOW users want responses (thinking preferences - brevity, detail, structure)

Installation

pip install langchain-prefid

Quick Start

from langchain_anthropic import ChatAnthropic
from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain_core.prompts import ChatPromptTemplate
from langchain_prefid import create_prefid_tools

# Create PrefID tools
tools = create_prefid_tools(
    client_id="your-prefid-client-id",
    access_token="user-access-token",
    user_id="user_123"
)

# Create agent
llm = ChatAnthropic(model="claude-3-5-sonnet-20241022")

prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful assistant. Use the preference tools to personalize responses."),
    ("placeholder", "{chat_history}"),
    ("human", "{input}"),
    ("placeholder", "{agent_scratchpad}"),
])

agent = create_tool_calling_agent(llm, tools, prompt)
executor = AgentExecutor(agent=agent, tools=tools)

# Use it!
result = executor.invoke({
    "input": "Recommend a restaurant for tonight"
})

Available Tools

1. get_user_preferences

Get user preferences for any domain (food, music, travel, coding, etc.)

# Agent will call this tool
"Get user's food preferences to recommend restaurants"

2. get_thinking_preferences

Get HOW the user wants AI to respond (Atom of Thought)

# Returns:
# - Reasoning style (stepwise, summary_first)
# - Decision style (recommend, tradeoffs, options)
# - Verbosity (concise, detailed, examples)
# - Autonomy (proactive, confirm_first)

3. learn_thinking_preference

Learn and remember how the user wants responses

# When user says: "I prefer brief answers"
# Agent calls this tool to remember it

4. explain_response_style

Introspection - explain why AI is responding a certain way

# When user asks: "Why are you responding like this?"
# Agent explains active preferences

Example Agent

from langchain_anthropic import ChatAnthropic
from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain_core.prompts import ChatPromptTemplate
from langchain_prefid import create_prefid_tools

# Setup
tools = create_prefid_tools(
    client_id="...",
    access_token="...",
    user_id="user_123"
)

llm = ChatAnthropic(model="claude-3-5-sonnet-20241022")

system_prompt = """You are a restaurant recommendation assistant.

BEFORE recommending:
1. Call get_thinking_preferences() to learn HOW to respond
2. Call get_user_preferences('food_profile') to learn WHAT they like
3. Structure response according to their preferences

If user asks why you responded that way, call explain_response_style().
If user tells you how they prefer responses, call learn_thinking_preference().
"""

prompt = ChatPromptTemplate.from_messages([
    ("system", system_prompt),
    ("placeholder", "{chat_history}"),
    ("human", "{input}"),
    ("placeholder", "{agent_scratchpad}"),
])

agent = create_tool_calling_agent(llm, tools, prompt)
executor = AgentExecutor(agent=agent, tools=tools, verbose=True)

# Run
response = executor.invoke({
    "input": "Where should I eat tonight?"
})

Individual Tools

You can also use tools individually:

from langchain_prefid import (
    PrefIDPreferenceTool,
    PrefIDThinkingTool,
    PrefIDLearnTool,
    PrefIDWhyTool
)
from prefid import PrefID

prefid = PrefID(client_id="...", access_token="...")

# Create individual tools
pref_tool = PrefIDPreferenceTool(prefid_client=prefid)
thinking_tool = PrefIDThinkingTool(prefid_client=prefid)
learn_tool = PrefIDLearnTool(prefid_client=prefid)
why_tool = PrefIDWhyTool(prefid_client=prefid)

# Use them
food_prefs = pref_tool.invoke({"domain": "food_profile"})
thinking = thinking_tool.invoke({"user_id": "user_123"})

Key Concepts

Atom of Thought (AoT)

PrefID's unique feature - learns HOW users want AI to respond:

• "I prefer step-by-step explanations" → Reasoning: stepwise
• "Just give me one recommendation" → Decision: recommend
• "Keep it brief" → Verbosity: concise

This goes beyond WHAT users like to HOW they think.

Thinking Preferences vs Content Preferences

• Content: What music/food/travel they like
• Thinking: How they want information structured

Both are important for truly personalized AI.

Examples

See /examples for complete working examples:

• restaurant_recommender.py - Full agent with all tools
• basic_agent.py - Simple preference-aware agent

LangSmith Hub

Standard prompts for PrefID agents are available on LangSmith Hub:

Prompt	Description
talentxmdu/restaurant-recommender	Full agent prompt with AoT + Preference logic
talentxmdu/basic-agent	Minimal prompt for preference aware agents

Usage:

from langchain import hub
prompt = hub.pull("talentxmdu/restaurant-recommender")

Get PrefID Credentials

1. Sign up at prefid.in
2. Create an OAuth client in Dashboard → Connected Apps
3. Get your client_id and client_secret

Links

• Documentation: https://docs.prefid.in
• PrefID SDK: https://github.com/Talentxmdu/PrefID
• LangChain: https://python.langchain.com/
• Support: hello@prefid.in

License

MIT © PrefID Team