coralbricks-langchain 0.3.0

CoralBricks + LangChain integration: retriever, chat history, memory helper, and agent tools.

coralbricks-langchain

Use CoralBricks as the memory backend for your LangChain applications: retriever, chat history, agent memory tools, and RAG middleware — all backed by the CoralBricks Memory API.

• Drop-in CoralBricksRetriever for any LCEL chain or RAG pipeline.
• Three agent tools (store, search, forget) for persistent memory across turns.
• CoralBricksChatMessageHistory — persistent chat history backed by the CoralBricks chat API.
• Memory stores — each store gets a dedicated index. Share memory across agents via the same store name.

---

Installation

pip install coralbricks-langchain

Requires Python 3.10+ and LangChain >= 1.0.

---

API key

Get a CoralBricks API key from the CoralBricks web app.

---

Quick start

from coralbricks_langchain import CoralBricksMemory

memory = CoralBricksMemory(api_key="your_coralbricks_api_key")
memory.get_or_create_memory_store("langchain:my-app")
memory.set_session_id("user-123")

# Save a memory
mem_id = memory.save_memory("Pro plan costs $199/month with unlimited ops.")

# Search by meaning
hits = memory.search_memory("What does the Pro plan cost?", top_k=3)
for h in hits:
    print(h.get("score"), h.get("text"))

# Forget by meaning
memory.forget_memory("Pro plan pricing")

---

Agent memory tools

Give your agent tools to store, search, and forget memories across turns. Pass the memory instance directly to get_tools() — no global state required.

from coralbricks_langchain import CoralBricksMemory, get_tools
from langchain_openai import ChatOpenAI

memory = CoralBricksMemory(api_key="your_coralbricks_api_key")
memory.get_or_create_memory_store("langchain:support-agent")
memory.set_session_id("user-123")

tools = get_tools(memory)

model = ChatOpenAI(model="gpt-4o-mini", temperature=0)
agent = create_agent(
    model,
    tools=tools,
    system_prompt=(
        "You are a helpful assistant with persistent memory (CoralBricks). "
        "Before answering, always search memory for relevant context. "
        "When you learn something important, store it."
    ),
)

result = agent.invoke({"messages": [{"role": "user", "content": "Remember: Alex is on the Enterprise plan."}]})
print(result["messages"][-1].content)

---

Retriever (LCEL chains)

CoralBricksRetriever implements BaseRetriever and drops into any LCEL chain:

from coralbricks_langchain import CoralBricksMemory, CoralBricksRetriever

memory = CoralBricksMemory(api_key="your_coralbricks_api_key")
memory.get_or_create_memory_store("langchain:my-kb")

retriever = CoralBricksRetriever(memory=memory, top_k=5)

docs = retriever.invoke("What is the cancellation policy?")
for doc in docs:
    print(doc.page_content)

---

Chat message history

Persistent chat history backed by the CoralBricks chat API:

from coralbricks_langchain import CoralBricksMemory, CoralBricksChatMessageHistory

memory = CoralBricksMemory(api_key="your_coralbricks_api_key")
history = CoralBricksChatMessageHistory(client=memory.client, conversation_id="conv-001")

history.add_user_message("Hello!")
history.add_ai_message("Hi, how can I help?")

for msg in history.messages:
    print(msg.type, msg.content)

---

API reference

CoralBricksMemory

Method	Description
CoralBricksMemory(api_key, base_url?)	Create memory instance (client created internally)
.get_or_create_memory_store(name)	Attach to or create a dedicated memory store (idempotent)
.create_memory_store(name)	Create a new store (raises if exists)
.set_project_id(id)	Set project namespace
.set_session_id(id)	Set session/user namespace
.save_memory(text, metadata?)	Embed and store a memory item
.search_memory(query, top_k=5)	Semantic search over memories
.forget_memory(query, top_k=5)	Forget memories matching a semantic query

Other components

Symbol	Description
CoralBricksRetriever	LangChain BaseRetriever for LCEL RAG pipelines
CoralBricksChatMessageHistory	LangChain BaseChatMessageHistory backed by CoralBricks chat storage
get_tools(memory)	Factory returning [store, search, forget] tools bound to a memory instance

---

Conventions

Field	Example	Purpose
store_name	langchain:support-agent	Dedicated index for this app/use-case
session_id	user-123, conv-001	Conversation or user scope
metadata	{"source": "policy"}	Optional metadata stored with each item

---

License

Apache-2.0.