Personalizing the agents using LLM

[spannercode]

Lets think about how we can use LLM to enhance the agents to make better reasoning or decisions.

Agents receive events from the chain or system and then makes a decision whether or not to act on them. For this, lets use the LLM.

Each agent has a system prompt that the admin (or owner of the agent can define; also default exists) that will guide the agent LLM to make decisions.

For the actions where we have probability, we can also have an option to select the LLM to make the decision to act or not.

[spannercode]

Lets first list down, where LLM usage makes most sense.

[mesudip]

Blocked by:

• Support metadata fetch for fields in transactions. #276

[Sital999]

LLM could be used for

• Creation of metadata for vote, proposal creation, drep registration.
• For voting on different type of proposals with different option with certain probability.
• Delegating stake to online dreps for once
• Creation of proposals on certain events on chain like vote
• Transferring ada as reward on certain events like vote, proposal creation