Agent Memory: User Preferences / Rules

[zrosenbauer]

Caution

this is super rough but wanted to get the idea out there and on paper

Overview

Create a low-level primitive that allows for configuration of user preferences that can be appended to system prompts, and in addition provide the ability to generate/detect preferences/rules.

This could be similar to: https://docs.cursor.com/context/rules

Proposal

import { UserPreference, Memory } from "@volt/core";

const userPreference = new UserPreference({
   // Optionally auto-generate preferences based on the conversation
   model: openai('gpt-4.1'),
   // OR 
   generate: async (...args) => {
       return myCustomGenerateCall(...args);
   }
});

// manually add a preference
userPreference.add({
  type: 'core',
  preference: 'Output all code examples as TypeScript'
});

const agent = new Agent({
  ...
  memory: new Memory({
    preferences: [userPreference] // allow for N number of preferences
  });
});

This would insert the preferences as context into the system prompt(s):

<user_rules>
- Output all code examples MUST be in TypeScript
- All diagrams MUST be in mermaid
- Only provide AWS related examples
</user_rules>

[omeraplak]

Hey @zrosenbauer ,
This is definitely something we want to support in VoltAgent. We could even build it as an adapter and offer support through VoltOps as well. I’ll take a closer look and see what we can do.

[zrosenbauer]

This could either build off of memory that already exists or be a completely new primitive. This could also be built ontop of the RAG implementation as it would sort of make sense to just grab that data every single time... IMO types and workflow are higher priority atm as this is a bigger project