OpenBB LLM Agents

To develop the project, use poetry for environment management and ruff for linting and formatting.

Setup Development Environment:

poetry install

Linting and Formatting with ruff:

• Run checks: ruff check
• Fix errors: ruff check --fix
• Format code: ruff format
• Install pre-commit hooks: pre-commit install

Testing: Run tests using pytest:

pytest -n 8 tests/

Install the openbb-agents package from PyPI. This project requires Python 3.10 or higher.

pip install openbb-agents --upgrade

To access financial data, you must provide credentials for the OpenBB Platform data providers. You can do this using one of two methods:

1. Local Configuration: Specify credentials in a ~/.openbb_platform/user_settings.json file.
2. OpenBB Hub: Use a Personal Access Token (PAT) from your OpenBB Hub account and pass it directly to the agent function.

OpenBB LLM Agents require an OpenAI API key to function. You must set the OPENAI_API_KEY environment variable in your shell profile (e.g., .bashrc or .zshrc).

export OPENAI_API_KEY="your_openai_api_key"

To use openbb-agents, you need to provide credentials for the OpenBB Platform data providers. You can fetch these using an OPENBB_PAT (Personal Access Token) from your environment variables, or set up API keys locally. If you rely only on yfinance, some functions may be unavailable.

import os
OPENBB_PAT = os.environ.get('OPENBB_PAT')

The agent can access OpenBB data through two methods:

1. OpenBB Hub (via PAT): Pass the openbb_pat argument to the agent function. This is the recommended way to manage credentials remotely.
2. Local Credentials: If openbb_pat is not provided, the agent will attempt to use credentials already configured in your local OpenBB environment.

When a openbb_pat is provided, the agent internally calls obb.account.login(pat=openbb_pat) to authenticate.

For asynchronous workflows, use the aopenbb_agent function.

from openbb_agents.agent import aopenbb_agent

# Async usage
await aopenbb_agent("What is the current market cap of TSLA?")

The configure_logging(verbose: bool) function sets up the global logging configuration for the openbb_agents package.

When verbose=True:

• openbb_agents.agent, openbb_agents.utils, openbb_agents.tools, and openbb_agents.chains log at the INFO level.
• The root logger logs at the WARNING level.

When verbose=False:

• All specified loggers and the root logger log at the CRITICAL level.

Logs are directed to sys.stdout using a standard formatter: %(asctime)s - %(levelname)s - %(name)s - %(message)s.

For asynchronous workflows, use aopenbb_agent. This can improve performance in certain scenarios by allowing the agent to run within an async event loop.

from openbb_agents.agent import aopenbb_agent

await aopenbb_agent("What is the stock price of AAPL and MSFT?")

The aopenbb_agent function is the asynchronous variant of openbb_agent. It is designed for use in asyncio environments, allowing for concurrent processing of subquestions to improve performance.

Parameters

• query (str): The question to be answered.
• openbb_tools (list[str] | None): A list of specific OpenBB function names to restrict the agent's toolset.
• openbb_pat (str | None): The OpenBB PAT for retrieving credentials from the OpenBB Hub.
• extra_tools (list[Callable] | None): A list of additional Python functions to add to the agent's tool index.
• verbose (bool): Whether to print additional information about the agent's behavior.