Adding semantic memory feature / embedding ?

[Finn-Chaosmedia]

With increasingly agentic applications, a major advantage is a memory for essential thoughts, facts, fundamentals, and processes.
Previously with AI, every chat started from scratch. The chat had to be filled with a foundation. The results from discussions in a chat can become better and more targeted when such information can be drawn upon.

Along with this, an embedding is helpful for quickly finding suitable information in data.

Even when using agent systems, an embedding feature is relevant that cannot be provided by mlx serve. Therefore, other tools must be run in parallel. This does not correspond to a central unit, but rather continues to be a patchwork of tools. mlx serve could still improve in this regard.