Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.langchain.com/llms.txt

Use this file to discover all available pages before exploring further.

This notebook shows how to build tools using the OpenGradient toolkit. This toolkit gives users the ability to create custom tools based on models and workflows on the OpenGradient network.

​
Setup

Ensure that you have an OpenGradient API key in order to access the OpenGradient network. If you already have an API key, simply set the environment variable: 
!export OPENGRADIENT_PRIVATE_KEY="your-api-key"
If you need to set up a new API key, download the opengradient SDK and follow the instructions to initialize a new configuration. 
!pip install opengradient
!opengradient config init

​
Installation

This toolkit lives in the langchain-opengradient package: 
pip install -qU langchain-opengradient

​
Instantiation

Now we can instantiate our toolkit with the API key from before. 
from langchain_opengradient import OpenGradientToolkit

toolkit = OpenGradientToolkit(
    # Not required if you have already set the environment variable OPENGRADIENT_PRIVATE_KEY
    private_key="your-api-key"
)

​
Build your own tools

The OpenGradientToolkit offers two main methods for creating custom tools:

​
1. Create a tool to run ML models

You can create tools that leverage ML models deployed on the OpenGradient model hub. User-created models can be uploaded, inferenced, and shared to the model hub through the OpenGradient SDK. 
import opengradient as og
from pydantic import BaseModel, Field


# Example 1: Simple tool with no input schema
def price_data_provider():
    """Function that provides input data to the model."""
    return {
        "open_high_low_close": [
            [2535.79, 2535.79, 2505.37, 2515.36],
            [2515.37, 2516.37, 2497.27, 2506.94],
            [2506.94, 2515, 2506.35, 2508.77],
            [2508.77, 2519, 2507.55, 2518.79],
            [2518.79, 2522.1, 2513.79, 2517.92],
            [2517.92, 2521.4, 2514.65, 2518.13],
            [2518.13, 2525.4, 2517.2, 2522.6],
            [2522.59, 2528.81, 2519.49, 2526.12],
            [2526.12, 2530, 2524.11, 2529.99],
            [2529.99, 2530.66, 2525.29, 2526],
        ]
    }


def format_volatility(inference_result):
    """Function that formats the model output."""
    return format(float(inference_result.model_output["Y"].item()), ".3%")


# Create the tool
volatility_tool = toolkit.create_run_model_tool(
    model_cid="QmRhcpDXfYCKsimTmJYrAVM4Bbvck59Zb2onj3MHv9Kw5N",
    tool_name="eth_volatility",
    model_input_provider=price_data_provider,
    model_output_formatter=format_volatility,
    tool_description="Generates volatility measurement for ETH/USDT trading pair",
    inference_mode=og.InferenceMode.VANILLA,
)


# Example 2: Tool with input schema from the agent
class TokenInputSchema(BaseModel):
    token: str = Field(description="Token name (ethereum or bitcoin)")


def token_data_provider(**inputs):
    """Dynamic function that changes behavior based on agent input."""
    token = inputs.get("token")
    if token == "bitcoin":
        return {"price_series": [100001.1, 100013.2, 100149.2, 99998.1]}
    else:  # ethereum
        return {"price_series": [2010.1, 2012.3, 2020.1, 2019.2]}


# Create the tool with schema
token_tool = toolkit.create_run_model_tool(
    model_cid="QmZdSfHWGJyzBiB2K98egzu3MypPcv4R1ASypUxwZ1MFUG",
    tool_name="token_volatility",
    model_input_provider=token_data_provider,
    model_output_formatter=lambda x: format(float(x.model_output["std"].item()), ".3%"),
    tool_input_schema=TokenInputSchema,
    tool_description="Measures return volatility for a specified token",
)

# Add tools to the toolkit
toolkit.add_tool(volatility_tool)
toolkit.add_tool(token_tool)

​
2. Create a tool to read workflow results

Read workflows are scheduled inferences that regularly run models stored on smart-contracts with live oracle data. More information on these can be found here. You can create tools that read results from workflow smart contracts: 
# Create a tool to read from a workflow
forecast_tool = toolkit.create_read_workflow_tool(
    workflow_contract_address="0x58826c6dc9A608238d9d57a65bDd50EcaE27FE99",
    tool_name="ETH_Price_Forecast",
    tool_description="Reads latest forecast for ETH price from deployed workflow",
    output_formatter=lambda x: f"Price change forecast: {format(float(x.numbers['regression_output'].item()), '.2%')}",
)

# Add the tool to the toolkit
toolkit.add_tool(forecast_tool)

​
Tools

Use the built in get_tools() method to view a list of the available tools within the OpenGradient toolkit. 
tools = toolkit.get_tools()

# View tools
for tool in tools:
    print(tool)

​
Use within an agent

Here’s how to use your OpenGradient tools with a LangChain agent: 
from langchain_openai import ChatOpenAI
from langchain.agents import create_agent


# Initialize LLM
model = ChatOpenAI(model="gpt-5.4")

# Create tools from the toolkit
tools = toolkit.get_tools()

# Create agent
agent_executor = create_agent(model, tools)

# Example query for the agent
example_query = "What's the current volatility of ETH?"

# Execute the agent
events = agent_executor.stream(
    {"messages": [("user", example_query)]},
    stream_mode="values",
)
for event in events:
    event["messages"][-1].pretty_print()
Here’s a sample output of everything put together: 
================================ Human Message =================================

What's the current volatility of ETH?
================================== Ai Message ==================================
Tool Calls:
  eth_volatility (chatcmpl-tool-d66ab9ee8f2c40e5a2634d90c7aeb17d)
 Call ID: chatcmpl-tool-d66ab9ee8f2c40e5a2634d90c7aeb17d
  Args:
================================= Tool Message =================================
Name: eth_volatility

0.038%
================================== Ai Message ==================================

The current volatility of the ETH/USDT trading pair is 0.038%.

​
API reference

See the GitHub page for more detail.
Connect these docs to Claude, VSCode, and more via MCP for real-time answers.
Edit this page on GitHub or file an issue.