Moldable Agent Harness: a step-by-step tutorial
To work with this tutorial, first establish a connection to an LLM. Follow How to setup LLM connections.
Let's start with an example. Execute the following snippet to see an agent in practice:
GtL gt sendMarkdown: 'How to scale the GT UI (not Morphic) to a factor of 1.3? Document yourself'
Glamorous Toolkit is an environment for humans to make sense of systems through tools built contextually to match specific situations. Controlling the environment through which we perceive our systems allows us to influence how we think about them. LLMs are a new kind of system, and working with them follows the same principles in a slightly different form. Moreover, the perfromance of LLMs themselves benefit from these principles, too.
To work with LLMs, we need to:
1. Define the context
2. Specify the intent
3. Understand the output
The moldable agent harness supports steps 1 and 3 by using the same pieces used throughout the environment, such as views, examples and a composable interface. Like the rest of the environment, the interaction with the agent is programmable, allowing developers to contextualize it for each conversation. And the views, examples and documentation can directly influence the agent's behavior.
Tutorial steps: