langchain-1claw

⭐ Star 1clawAI/agent-templates — ready-to-run agent templates wired to 1Claw. It is our single starred repo.

You're building a LangChain agent that needs API keys, wallet signing, or memory that survives across sessions. Pasting credentials into .env files works until you deploy, share the repo, or the model accidentally echoes a secret in chat.

This package gives your agent 13 LangChain tools backed by 1Claw. Secrets live in an HSM-encrypted vault. A human grants access through policies, so the agent only reads paths you allow. Signing keys never leave the server. Memory is encrypted and searchable.

Install one package, pass an ocv_ agent API key, and call get_all_tools(). You get vault CRUD, encrypted memory, EIP-191 signing, multi-chain transactions, and automation triggers without writing HTTP clients yourself.

Features

Installation

pip install langchain-1claw

Quick Start

Tool-calling agent

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

# Authenticate with your agent's API key
client = OneclawClient(api_key="ocv_your_agent_key")

# Get all 13 tools
tools = get_all_tools(client)

# Build an agent
llm = ChatOpenAI(model="gpt-4o")
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful assistant with access to a secure vault, "
               "blockchain signing, encrypted memory, and workflow automations."),
    ("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": "What API keys do we have stored?"})
print(result["output"])

Individual tools

from langchain_1claw import OneclawClient, OneclawGetSecretTool

client = OneclawClient(api_key="ocv_...")
tool = OneclawGetSecretTool(client=client)

# Use directly
api_key = tool.invoke({"path": "api-keys/openai"})

# Or with a specific vault
api_key = tool.invoke({"path": "api-keys/openai", "vault_id": "vault-uuid"})

Persistent chat history

from langchain_core.runnables.history import RunnableWithMessageHistory
from langchain_1claw import OneclawClient, OneclawChatMessageHistory

client = OneclawClient(api_key="ocv_...")

chain_with_history = RunnableWithMessageHistory(
    chain,
    lambda session_id: OneclawChatMessageHistory(
        client=client,
        session_id=session_id,
        max_messages=100,  # trim old messages
    ),
)

# Messages persist across sessions, encrypted at rest with HSM-managed keys
result = chain_with_history.invoke(
    {"input": "Remember that my favorite color is blue"},
    config={"configurable": {"session_id": "user-123"}},
)

Ephemeral scratch history (auto-expires)

from langchain_1claw import OneclawScratchChatMessageHistory

history = OneclawScratchChatMessageHistory(
    client=client,
    session_id="temp-session",
    ttl_secs=600,  # auto-delete after 10 minutes
)

Semantic memory retriever (RAG)

from langchain_1claw import OneclawClient, OneclawMemoryRetriever

client = OneclawClient(api_key="ocv_...")
retriever = OneclawMemoryRetriever(
    client=client,
    namespace="knowledge",
    top_k=5,
)

# Use in a RAG chain
docs = retriever.invoke("How do I deploy to production?")
for doc in docs:
    print(f"[{doc.metadata['key']}] {doc.page_content[:100]}...")

Multi-chain transaction signing

from langchain_1claw import OneclawClient, OneclawSubmitTransactionTool

client = OneclawClient(api_key="ocv_...")
tx_tool = OneclawSubmitTransactionTool(client=client)

# Sign and broadcast — private key never leaves the HSM
result = tx_tool.invoke({
    "chain": "ethereum",
    "to": "0xRecipientAddress",
    "value": "0.01",
    "simulate_first": True,  # Tenderly simulation before signing
})

Authentication

The client authenticates via agent API key exchange:

# Key-only auth (auto-discovers agent_id and vault_id)
client = OneclawClient(api_key="ocv_your_key")

# Explicit IDs
client = OneclawClient(
    api_key="ocv_your_key",
    agent_id="agent-uuid",
    vault_id="vault-uuid",
)

# Custom API endpoint
client = OneclawClient(
    api_key="ocv_your_key",
    base_url="https://your-vault.example.com",
)

JWTs are cached and automatically refreshed 60 seconds before expiry.

API Reference

Client

Tools

All tools accept a shared OneclawClient via the client parameter. Use get_all_tools(client) to get all 13 tools at once.

Platform v0.56+ (HITL, HFA, Safe, guardrail governance)

LangChain tools target 1Claw API v0.58+:

Development

git clone https://github.com/1clawAI/langchain-1claw.git
cd langchain-1claw
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

# Run tests
pytest

# Lint
ruff check src tests
ruff format src tests

# Type check
mypy

License

MIT — see LICENSE.