Warning
FastAgent is currently under active development and in alpha stage. It currently lacks SSL and CORS security features, as well as comprehensive testing needed for production deployment.
You can install the project using pip: (Available on PyPI soon)
pip install git+https://github.com/bastienpo/fastagent.gitFirst, create a simple LangChain application.
Let's create an app.py file with the simplest langchain runnable you can make, consisting of just a large language model.
# pip install -qU langchain-mistralai and requires MISTRAL_API_KEY in to be set
from langchain_mistralai import ChatMistralAI
chain = ChatMistralAI(model="ministral-3b-latest")Then, you need to initialize a fastagent configuration file (fastagent.toml) using the following command:
fastagent initYou will need to update the app field in the project section to match the path of your application in the form:
<module_path>:<module_attribute> in you case it would be myapplication.app:chain
If you decided to use a database in your configuration, you can use the setup command to create the tables and setup the database.
fastagent setupWhen you are ready to develop or ship, you can start the production server:
fastagent dev # fastagent runThe difference between the dev and run command is that dev will reload the server on code changes and log to the console, while run will build a docker image and serve the application.
This is the current roadmap for the project:
- The project allows to setup a simple server for your agent. It only support langchain and postgresql to store your conversations and authentication data.
- It allows to setup a simple token based authentication and authorization.
- You can configure different middlewares CORS, Authorization and Rate limiting.
Note
This project is still under active development and design is subject to change. I'm doing it only as a personal project because I was curious about how to ship an agent to production. Feel free to contribute or give feedback. I'm really open to any suggestion.
Near future:
- Add end-to-end testing of the project for langchain and langgraph.
- Add a permission system.
- Add stateless authentication.
- Add support for more databases (SQLite in particular).
Long term:
- Simplify the integration of monitoring and observability tools (Langfuse, Phoenix)
- Add support for Dspy and other frameworks.
Some resources about the dependencies used for the project and thanks to the maintainers of the projects for their work.
The project is inspired by LitServe.