Back to Home
Open Source

XiaoyaoClaw Memory Distill

Distill chats into structured memory and fix context overflow — semantically tiered writes to MEMORY.md and daily logs, incremental dedup, sensitive-data guard. Agents never lose memory.

Product Introduction

OpenClaw Memory Distill solves four memory-management pains at once: context overflow, fragmented memory, missing MEMORY.md, and bloat. OpenClaw sessions start fresh — long chats get key info pushed out, forcing /reset. This skill distills chats via semantic tiering (core → MEMORY.md at root, daily → memory/YYYY-MM-DD.md, ephemeral → day-only); auto-builds the first MEMORY.md by scanning history when missing; incremental dedup before writes (append new, merge duplicates); auto-detects tokens/passwords/keys and never writes them by default; expired logs archive to archive/, never auto-delete. Three triggers — one-shot manual, daily Cron, dynamic HEARTBEAT. Complements the workspace initializer: it manages the "home", this one manages the "content".

XiaoyaoClaw Memory Distill poster

Core Features

  • Semantic tiering — core → MEMORY.md, daily → day log, ephemeral stays day-only (semantic, not keyword)
  • First-time memory — scan history to auto-derive a starter MEMORY.md
  • Incremental dedup — pre-write check, append new, merge duplicates
  • Sensitive-data guard — auto-detect tokens / passwords / keys, never persist by default
  • Archive-only — expired logs go to archive/, deletion needs human confirmation
  • Three triggers — one-liner / daily Cron / HEARTBEAT
  • Multi-agent isolation — each agent only touches its own workspace
  • Pairs with initializer — paths follow WORKSPACE.md, config-safety inherited
  • Easy install — ClawHub one-liner (clawhub install xiaoyaoclaw-memory-distill) or manual from GitHub

Use Cases

OpenClaw / XiaoyaoClaw usersLong-session / context-overflow agentsMulti-agent independent memoryLong-term agent knowledge and decision logging

Tags

#OpenClaw#AI Skill#Memory mgmt#Context mgmt#Multi-agent#MIT