kyma-memorylisted

# Kyma — the user's context engine This project is wired to **Kyma**: the user's durable memory **+** live data **+** the knowledge graph that links them. The `kyma-memory` plugin bundles an MCP server named `kyma` (memory + data + graph tools) and **hooks** that automatically capture each session and inject the most relevant memories into every prompt — so recall is already happening; use the tools below when you need *more*, or to write. ## When to recall Call `recall_memory` (MCP server `kyma`) before answering anything that could depend on prior context: the user's preferences, past decisions, project conventions, architecture choices, or "how we did X last time". Pass the user's request as `query`, set `realms` to the current project (the working directory's basename) plus `global`, and `limit` ~8. Recalled memories also arrive automatically as injected context on each prompt — use `recall_memory` when you need more, a different query, or a specific realm. ## When to save Call `save_memory` when the user states something durable and reusable: - **decision** — an architecture/approach choice and its rationale - **preference** — how the user likes things done (style, tools, workflow) - **fact** — a non-obvious, load-bearing fact about the system or domain - **learning** — something discovered this session worth keeping Set `realm` to the project (or `global` for cross-project truths) and `importance` 0.3–0.9. Keep each memory self-contained (one idea, understandable