langchain-thrindex 0.1.0

Official LangChain / LangGraph integration for Thrindex — persistent semantic memory for AI agents.

langchain-thrindex

Official LangChain / LangGraph integration for Thrindex — persistent, semantically-searchable long-term memory for AI agents.

Installation

pip install langchain-thrindex

Components

Class	Interface	Use case
ThrindexStore	langgraph.store.base.BaseStore	LangGraph agents — long-term memory across threads
ThrindexRetriever	langchain_core.retrievers.BaseRetriever	LCEL chains, RAG pipelines

---

ThrindexStore — LangGraph agents

The primary integration. Pass ThrindexStore to graph.compile(store=store) and your LangGraph agent gets persistent memory that survives restarts and scales across deployments.

from langchain_thrindex import ThrindexStore
from langgraph.prebuilt import create_react_agent

store = ThrindexStore(
    api_key="th_live_...",
    agent_id="customer-support-bot",  # stable identifier for your agent
)

# Compile the agent with persistent memory
agent = create_react_agent(
    model="openai:gpt-4o",
    tools=[...],
    store=store,
)

# The agent now reads and writes long-term memories automatically
result = agent.invoke(
    {"messages": [{"role": "user", "content": "My name is Alice and I prefer dark mode."}]},
    config={"configurable": {"thread_id": "thread-1"}},
)

Manual store operations

# Store a fact
store.put(
    ("memories", "user-42"),  # namespace: (category, user_id)
    "preference-display",     # unique key within the namespace
    {"content": "User prefers dark mode and compact layout"},
)

# Semantic search
results = store.search(
    ("memories", "user-42"),
    query="display preferences",
    limit=5,
)
for item in results:
    print(item.value["content"], item.score)

# Get by key
item = store.get(("memories", "user-42"), "preference-display")

# Delete
store.delete(("memories", "user-42"), "preference-display")

Async

All operations have async counterparts:

await store.aput(namespace, key, value)
item = await store.aget(namespace, key)
results = await store.asearch(namespace, query="...")
await store.adelete(namespace, key)

Namespace → Thrindex mapping

LangGraph	Thrindex
agent_id constructor arg	agent_id in every API call
"/".join(namespace)	user_id in every API call

---

ThrindexRetriever — LCEL chains

from langchain_thrindex import ThrindexRetriever
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI

retriever = ThrindexRetriever(
    api_key="th_live_...",
    agent_id="my-agent",
    user_id="user-42",
    k=5,
    task_context="answering user questions about product settings",
)

prompt = ChatPromptTemplate.from_template(
    "Use the following memories to answer the question.\n\n"
    "Memories:\n{context}\n\n"
    "Question: {question}"
)

chain = (
    {"context": retriever, "question": lambda x: x}
    | prompt
    | ChatOpenAI(model="gpt-4o")
    | StrOutputParser()
)

answer = chain.invoke("What display settings does the user prefer?")

---

Configuration reference

ThrindexStore

Parameter	Type	Default	Description
api_key	str	required	Thrindex API key
agent_id	str	"default"	Stable agent identifier
base_url	str	"https://api.thrindex.com"	API base URL (override for self-hosted)
timeout	float	30.0	Request timeout in seconds
max_retries	int	3	Retries on transient failures
extract	bool	True	Run LLM fact extraction on stored memories

ThrindexRetriever

Accepts api_key, agent_id, user_id, k, task_context, base_url, timeout, max_retries. See class docstring for full details.

---

Links

• Thrindex docs
• LangGraph store guide
• GitHub