Feature: Optional external memory backend for persistent context across sessions

[ferhimedamine]

KoboldCpp is excellent for local inference but sessions are ephemeral — context is lost on restart and is limited by the context window. This proposes optional Dakera (https://dakera.ai) integration as an external memory backend that persists context across sessions.

Problem: KoboldCpp users running persistent companions or long-form creative writing lose all context when the context window fills or the server restarts.

Proposed: Optional --memory-url flag:
koboldcpp.py --model mymodel.gguf --memory-url http://localhost:3300 --memory-key demo

At generation time:

1. Key events/exchanges are stored via POST /v1/memories
2. Before generation, relevant past context is retrieved via POST /v1/memories/search and prepended to the story

Why Dakera fits KoboldCpp's philosophy:

• Both are self-hosted first — no cloud dependencies
• Both run on consumer hardware
• Dakera's decay weighting is narratively appropriate: recent events are vivid, old ones fade

Setup: docker run -d -p 3300:3300 -e DAKERA_API_KEY=demo ghcr.io/dakera-ai/dakera:latest (or native binary available)

Happy to open a PR implementing this as an optional extension.

[LostRuins]

Koboldcpp already has its own inbuilt memory system, that's saved in browser or on exported json saves, and certainly not "lost". It also has embedding search and textdb