langchain-opengradient

This package contains the LangChain integration with OpenGradient.

More information about OpenGradient can be found here.

Installation

pip install -U langchain-opengradient

And you should configure credentials by setting the following environment variables:

OPENGRADIENT_PRIVATE_KEY - Your OpenGradient private API key

If you do not have an OpenGradient private key configured you can get one by running

pip install opengradient
opengradient config init

Toolkits

OpenGradientToolkit class provides a set of functions for creating tools that integrate OpenGradient models and workflows into LangChain agents.

from langchain_opengradient import OpenGradientToolkit
import opengradient as og
from pydantic import BaseModel, Field
from typing import List

# Initialize the toolkit
# Either set the environment variable "OPENGRADIENT_PRIVATE_KEY"
# or directly pass in private key.
toolkit = OpenGradientToolkit(private_key="MY_PRIVATE_KEY")

# Example 1: Create a volatility prediction tool with no input schema
def model_input_provider():
    return {
        "open_high_low_close": [
            [2535.79, 2535.79, 2505.37, 2515.36],
            [2515.37, 2516.37, 2497.27, 2506.94],
            # ... more price data
        ]
    }
    
def output_formatter(inference_result):
    return format(float(inference_result.model_output["Y"].item()), ".3%")
    
volatility_tool = toolkit.create_run_model_tool(
    model_cid="QmRhcpDXfYCKsimTmJYrAVM4Bbvck59Zb2onj3MHv9Kw5N",
    tool_name="eth_usdt_volatility",
    model_input_provider=model_input_provider,
    model_output_formatter=output_formatter,
    tool_description="Generates volatility measurement for ETH/USDT",
    inference_mode=og.InferenceMode.VANILLA,
)

# Example 2: Create a tool with an input schema
class VolatilityInputSchema(BaseModel):
    token: str = Field(description="Token name (e.g., 'ethereum' or 'bitcoin')")

def model_input_provider_with_schema(**llm_input):
    token = llm_input.get("token")
    # Fetch appropriate data based on token
    if token == "bitcoin":
        return {"price_series": [100001.1, 100013.2, 100149.2, 99998.1]}    # Replace with live data
    elif token == "ethereum":
        return {"price_series": [2010.1, 2012.3, 2020.1, 2019.2]}           # Replace with live data
    else:  # ethereum
        raise ValueError("Received unexpected token")

token_volatility_tool = toolkit.create_run_model_tool(
    model_cid="QmZdSfHWGJyzBiB2K98egzu3MypPcv4R1ASypUxwZ1MFUG",
    tool_name="token_volatility",
    model_input_provider=model_input_provider_with_schema,
    model_output_formatter=lambda x: format(float(x.model_output["std"].item()), ".3%"),
    tool_input_schema=VolatilityInputSchema,
    tool_description="Measures return volatility for specified token"
)

# Example 3: Create a workflow reading tool
workflow_tool = toolkit.create_read_workflow_tool(
    workflow_contract_address="0x58826c6dc9A608238d9d57a65bDd50EcaE27FE99",
    tool_name="ETH_Price_Forecast",
    tool_description="Reads latest forecast for ETH price",
    output_formatter=lambda x: f"Price change forecast: {
        format(float(x.numbers['regression_output'].item()), '.2%')
        }"
)

# Add tools to the toolkit
toolkit.add_tool(volatility_tool)
toolkit.add_tool(token_volatility_tool)
toolkit.add_tool(workflow_tool)

# Get all tools
tools = toolkit.get_tools()

# Use with an agent
from langgraph.prebuilt import create_react_agent
from langchain_openai import ChatOpenAI

llm = ChatOpenAI()
agent_executor = create_react_agent(llm, tools)

example_query ="What's the current volatility of ETH/USDT?"

events = agent_executor.stream(
    {"messages": [("user", example_query)]},
    stream_mode="values",
    )

for event in events:
    event["messages"][-1].pretty_print()