Skip to main content

FOR EVERYONE — OPTIONAL

Memory

By default, each session starts fresh. That's fine for quick work — but on anything that runs over days, you don't want to keep re-explaining what you decided last time. Turn on memory and the team remembers.

This is the kit's Memoria value made real: context that survives.

What it gives you​

  • Decisions that stick. When the team makes a call — an architecture choice, a brand decision, a "we're not doing X" — it can record it, and recall it later.
  • No re-explaining. Start a fresh session next week and the relevant context is already there.
  • A clean close. When you wrap up a session, the team can write a short, curated summary of what happened — so the next session opens with a grounded picture, not a blank page.

Turning it on​

Memory is off by default. You opt in at setup:

  • Answer "yes" to memory when /kit-init asks, or
  • Run with --memory on.

During setup, the kit offers to connect the memory backend and explains it first. If the backend isn't installed yet, memory stays quietly dormant — it never errors — until it's present.

Per-project "wings"​

Each project maps to its own memory wing — a separate room for that project's context. This is why the kit is per-project: your client work and your side project never bleed into each other's memory. Decisions, plans, and the team's working notes all stay filed under the right project.

What it's not​

Memory doesn't watch you or phone home. It's a per-project record the team keeps for you, stored through the memory backend you connected, and it stays off entirely unless you turn it on.

How it works under the hood
  • Backend. Memory is wired to MemPalace, registered as an MCP server during opt-in setup (or you're guided to install the mempalace-mcp runtime). Until it's present, the memory rule is a safe no-op — the team behaves exactly as if memory were off.
  • The hooks. When enabled, /kit-init resolves the IF:MEMORY blocks and installs session hooks — SessionStart (recall context for this wing), Stop (record decisions / write the close summary), and PreCompact (preserve context before the conversation is compacted).
  • The rule. A mempalace rule teaches the team when to save (decisions worth keeping, filed to the right "room") and when to recall (at the start of relevant work). On session close it writes a curated summary rather than dumping the raw transcript.
  • Config. kit.config.json carries .memory — whether it's on and the project's wing — surfaced to scripts as KIT_MEMORY and KIT_WING.

Where to go next​