Agent FRED

Documentation: https://ChrisPappalardo.github.io/agent_fred

Source Code: https://github.com/ChrisPappalardo/agent_fred

---

Agentic LLM interface for Federal Reserve Economic Data ("FRED") using a locally-hosted open-source Large Language Model.

FRED is hosted and maintained by the St. Louis Federal Reserve Bank. Please see the FRED Terms of Use for permissible use, copyright and other legal restrictions, and additional information.

Prerequisites

This project assumes you have Linux or Windows running WSL2. I have not tested it with macOS. You do not need a GPU to run this application.

This python app requires ollama to host and serve LLM API. Select your platform and follow the ollama installation directions.

You will also need to download whatever model(s) you intend to use. The default is Microsoft's 3B phi3 LLM for both the agent and chat pipelines. After installing ollama you can install phi3 with:

ollama pull phi3

You will need Poetry and at least python 3.10.

Installation

Clone this repository:

git clone https://github.com/ChrisPappalardo/agent_fred.git

Create a poetry shell inside the directory and install the requirements:

cd agent_fred
poetry shell
poetry install

Quickstart

Configuration

This python app uses environment variables to configure the AI backend and other variables. See src/agent_fred/config.py.

If you use the phi3 LLM you don't need to set anything but the FRED_API_KEY, which you can obtain here:

export FRED_API_KEY=1234

You can also set your config environment in .env in the project root:

FRED_API_KEY=123

The configurable options are as follows:

• FRED_API_KEY API key to use when making calls to the FRED
• FRED_LLM LLM to use with Agent FRED (default=phi3)
• FRED_TEMPERATURE verbosity of the model (default=0.0)

Run

Agent FRED has both a command-line interface and a web application.

To use the web application (recommended):

python app.py

Navigate to http://127.0.0.1:7860 to view the app:

You can also use the CLI, which returns API results in JSON:

python src/agent_fred/chat.py

Usage

It typically helps to have an understanding of the data available from the FRED API.

An example script you can use to test the application:

• What data is available from the FRED?
• What is GDP and how is it calculated?
• What was GDP from 2000-01-01 through 2023-12-31?
• Why did it dip in 2020?

In the web app you can clear the LLM's memory and start a new Q&A session with the Clear button.

Enjoy!

Testing

If you're developing with Agent FRED, you can run the unit tests using the Makefile like so:

make test

You can run the linters using pre-commit:

make pre-commit

Documentation

The documentation is automatically generated from the content of the docs directory and from the docstrings of the public signatures of the source code. The documentation is updated and published as a Github Pages page automatically as part each release.

---