Skip to main content

Development Mode in Agenta

Description: This quick start guide provides pointers for contributing to Agenta by working on the CLI, SDK, backend, and frontend code. Additionally, it covers how to debug the backend effectively.

Development mode for CLI or SDK​

If you are making changes to the CLI or SDK, you certainly want to use your version of the code instead of the one that you installed using pip. Here is how you can do that

  1. Navigate to the agenta-cli folder.
  2. Activate the virtual environment by running poetry shell in your terminal.

Now, you should be using your own version of agenta. To make sure that is true, run which agenta and check that the folder that you get is the one from agenta-cli.

If running agenta in your terminal still doesn't work after following the above steps, try the following:

Solution 1:

  1. Leave the poetry environement (simply run exit)
  2. Run pip uninstall agenta
  3. Run again poetry shell

Don't forget to reinstall agenta after you finish development.

Solution 2:

  • In the terminal where the poetry virtual environment is activated, run which agenta.
  • Copy the path where agenta is located, and add export agenta=/replace/path/to/agenta at the last line in your shell configuration file. If you're running Ubuntu, your shell configuration file will be ~/.bashrc.
  • Go to the terminal where you tried running agenta, and apply the changes you've made by running source ~/.bashrc. Proceed to run export to confirm agenta has been exported.
  • If you see agenta along with its path, proceed to initialize your application by running $agenta init and follow through the steps.
  • When you're ready to serve your app, run $agenta variant serve.

How to streamline working on the SDK​

Let's say you want to add a new type of parameters to the SDK, let's say IntParam, how can you work and test that quickly? Here is how I do it:

  • Create a new application, a new variant, and serve it. This will copy the agenta folder into the application folder
  • Go to main.py, it should look like this:

if __name__ == "__main__":
    assert os.environ["ROOT_PATH"] != ""
    run("agenta:app", host="0.0.0.0", port=80, root_path=os.environ["ROOT_PATH"])

Change the main.py to this

if __name__ == "__main__":
    run("agenta:app", host="0.0.0.0", port=801, auto_reload=True)

Rename the file including the code of your variant to _app.py

Now run python main.py (making sure of course that you have dealt with the dependencies), this will start uvicorn and publish the api of the variant to the port 801

Now you can make changes to the code of the sdk lying in the subfolder agenta and the code of your app and see the changes directly in the api. So, for our example, I would add the class for IntParam to the SDK, use it in the app, then take a look at the openapi.json to see if it works correctly, then run the app using the shell to see if everything works correctly.

Working on Backend Code​

No special steps are required for working on the backend code. Docker Compose automatically mounts the volume containing the code. Any changes you make will be immediately reflected in the backend.

Working on Frontend Code​

Similar to the backend, Docker Compose mounts the volume containing the frontend code. Consequently, any changes you make will be directly applied to the frontend.

It was noticed that sometimes when agenta-web is run as a standalone NextJS app and not as a Docker container, it takes lesser CPU usage. To run agenta-web as a standalone app in parallel with docker-compose, execute the start_web_standalone.sh script present in the project's root.

Debugging Backend​

The recommended method for debugging the backend is to check the logs from Docker. Here's how:

  1. Open Docker Desktop.
  2. Select the backend container.
  3. Click on the "Logs" button to access the logs. OR
  4. If you are using Visual Studio Code (VSCode), follow these steps:
    • Go to the Docker tab in VSCode.
    • Select the backend container.
    • Click on the "Logs" button to view the logs.

With these guidelines, you can efficiently contribute to Agenta by making local code changes and effectively debugging the backend. Happy coding!