Running the Agent¶
Requirements¶
You can choose between the following model backends for running your agents:
- Ollama installation instructions
- GitHub Copilot. You will need the Copilot CLI installed and configured.
See here for instructions on switching between backends
Note: We're working on adding support for more models and providers in the future. If you have a specific model or provider you'd like to see supported, please let us know by creating an issue in our GitHub repository.
Your first agent template¶
Create a file named hello.mgx:
---
description: Hello world agent template
---
<<
# Hello World
Tell the user Hello, and welcome them to Margarita!
>>
@effect run
Run it with:
margarita run hello.mgx
The << >> block loads markdown content into the agent's context. @effect run tells the agent to execute with the current context.
Specify the model¶
Add a model field to the template metadata:
---
model: "gpt-4"
---
<< test >>
@effect run
Naming the Runs¶
If you supply a parameter after @effect run, that parameter will be used as the name of the run.
---
description: Hello world agent template
---
<<
# Hello World
Tell the user Hello, and welcome them to Margarita!
>>
@effect run Hello World