LangGraph Plugin

Drop-in memory tools for LangGraph agent workflows. Provides @tool-decorated functions for use as LangGraph nodes or LLM tool-calling.

Requirements

• Python 3.9+

• langgraph >= 0.2

• langchain-core >= 0.3

Install

pip install tinyhuman-langgraph

Quick Start: Factory Pattern (Recommended)

Use make_memory_tools to create tools with credentials baked in. Credentials are never exposed to the LLM as tool parameters, preventing prompt-injection attacks.

from langchain_openai import ChatOpenAI
from tinyhuman_langgraph import make_memory_tools

tools = make_memory_tools(token="your-api-key")

# Bind to a model for LLM tool-calling
model = ChatOpenAI(model="gpt-4o").bind_tools(tools)

Usage as a LangGraph Node

from langgraph.graph import StateGraph, MessagesState
from tinyhuman_langgraph import make_memory_tools

ingest_tool, read_tool, delete_tool = make_memory_tools(token="your-api-key")

def memory_node(state: MessagesState):
    result = ingest_tool.invoke({
        "items": [{"key": "fact-1", "content": "User likes Python"}]
    })
    return {"messages": [f"Memory ingested: {result}"]}

graph = StateGraph(MessagesState)
graph.add_node("memory", memory_node)

Environment Variables

You can configure via environment instead of passing credentials directly:

export TINYHUMANS_API_KEY="your-api-key"

from tinyhuman_langgraph import get_tools

tools = get_tools()

Available Tools

Tool	Description
tinyhuman_ingest_memory	Upsert one or more memory items
tinyhuman_read_memory	Read items filtered by key / keys / namespace
tinyhuman_delete_memory	Delete items by key / keys / delete_all

Error Handling

Tools raise TinyHumanError on API failures and ValueError on invalid input. Both propagate normally through LangGraph.