Skip to main content

LangChain

If your LangChain agent needs API keys, wallet signing, or memory that outlives a single chat session, you have two bad defaults: paste secrets into .env and hope nothing leaks, or hand-roll HTTP calls against the 1Claw API in every project.

langchain-1claw is the official package. It wraps the agent API as LangChain tools, a chat message history, and a memory retriever. You pass an ocv_ agent key, call get_all_tools(), and wire the result into LangGraph or a tool-calling agent. Secrets stay in the vault. Signing happens server-side. Memory is encrypted and searchable.

Install​

pip install langchain-1claw

Prerequisites​

  1. A 1Claw account with a vault and at least one secret path your agent can read.
  2. An agent registered in your org with an ocv_ API key.
  3. Access policies that grant the agent read (and write, if needed) on the paths you expect the tools to touch.

Quick start​

from langchain_openai import ChatOpenAI
from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain_core.prompts import ChatPromptTemplate

from langchain_1claw import OneclawClient, get_all_tools

client = OneclawClient(api_key="ocv_your_agent_key")
tools = get_all_tools(client)

llm = ChatOpenAI(model="gpt-4o")
prompt = ChatPromptTemplate.from_messages([
    ("system", "You have access to a secure vault, signing keys, and encrypted memory."),
    ("human", "{input}"),
    ("placeholder", "{agent_scratchpad}"),
])

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

result = executor.invoke({"input": "List the API keys we have stored."})
print(result["output"])

agent_id and vault_id are optional. When omitted, the client resolves them from the token exchange response.

What you get​

CategoryTools
Secretsget, put, list, rotate
Memoryput, get, semantic search
SigningEIP-191 message sign, multi-chain transactions, balance check
Automationstrigger workflow runs

Plus OneclawChatMessageHistory for durable conversation storage and OneclawMemoryRetriever for RAG over agent memory.

MCP alternative​

If you already run LangChain with MCP adapters, you can point at the hosted 1Claw MCP server instead of installing this package. That path auto-discovers a larger tool set but adds a network hop. See Agent frameworks for MCP setup.

For most Python LangChain projects, langchain-1claw is simpler: typed tools, no MCP server to run, and the same policy gates on the backend.