Inject memory summary into chat context

[RocketJannis]

Hi, thanks for this plugin. I had the idea that the plugin could generate a short summary from memories from time to time and inject it as context into all chats similar to how the Claude's memory feature works.

So what's needed:

1. memory summary generation
2. summary injection

But idk if this is even possible with the current LM Studio API as the plugin would need to execute prompts by itself for the summary generation.

[RocketJannis]

A I see there is already something implemented in this direction but not mentioned in the readme and not sure if it works 🤔

https://github.com/imezx/persistent-memory-plugin/blob/master/src/preprocessor.ts

[imezx]

The current implementation in preprocessor.ts behaves a bit differently than a global, Claude-style profile summary mechanism.

How it works right now (Dynamic RAG)
instead of generating a high-level chronological summary, the plugin currently uses like a query-based RAG approach:

1. When you type a message, the preprocessor extracts key terms.
2. It queries the local database to pull the top-N most relevant individual memories based on semantic relevance and a time-decay factor.
3. It formats just those specific scattered facts and prepends them to your prompt (which you can see in the formatMemories function).

the plugin already makes asynchronous, out-of-band LLM calls behind the scenes to analyze your messages and extract facts. because the plugin environment gives us access to the underlying inference engine, we can trigger the LLM to run custom prompts whenever we want.