Claude Code Memory 2.0 (AutoDream)

Claude Code Memory 2.0 (AutoDream)

A three-layer memory system in Claude Code: normal session chat, Auto Memory (a memory.md file that stores per-project facts and preferences injected into every new session), and AutoDream. AutoDream runs a background sub-agent periodically that consolidates, prunes, and refreshes memory files — modelled explicitly on human sleep. Activated via /memory, runs via /dream or on a time/session interval, and only touches memory files (never code). Reduces repetition, bloat, and context rot across long-lived projects.

Key points

  • Three layers: session chat, Auto Memory (persistent memory.md), AutoDream (background cleanup)
  • AutoDream runs a sub-agent that consolidates, prunes, and rewrites memory files
  • Triggered by interval (hours) or session count — runs in background with a 'dreaming' status
  • Only touches memory files (.md), never code or scripts
  • Explicitly modelled on how humans consolidate memory during sleep
  • Activated globally across projects via /memory interface
  • Shows diff of added entries and removed sections after each dream
  • Jack Roberts describes a complementary manual architecture for Claude/Codex-style work: stable identity instructions, per-project operating manuals, and long-term Pinecone or Obsidian archives [src-059].
  • Roberts's emphasis is cross-app memory rather than only Claude Code internals: the outcome of a conversation should not depend on one chat history if the important context has been written into files or indexes [src-059].
  • In the Hermes Agent OS demo, Roberts extends Claude Code memory outward: Hermes can ask what the Claude/Codex-style operating system knows, read overnight reflections, and surface active work context through Telegram [src-079].

Related entities

Related concepts

Source references

  • [src-004] Nate Herk cluster — Nate Herk — Claude Code cluster (21 videos)

– Videos referenced: LrgfmZkl3nc, tXtCK66fPj8

  • [src-059] Jack Roberts — "This Memory System just 10x'd Claude Code" (2026-05-03)
  • [src-079] Jack Roberts — "Hermes Agent just got 10X Better (Agentic OS)" (2026-05-15)

Robin Cartier perspective

This page is part of Robin Cartier's working AI knowledge graph: a practical research layer for production AI, recommendation systems, experimentation, GEO, and agentic web readiness.

The useful next step is to connect this concept back to applied product leadership and operating models.

Recommended next

Keep reading from this thread

From 491 indexed pages and articles.

  1. Wiki concept Three-Layer AI Memory Roberts's structure for giving an assistant durable context: short-term memory answers "who am I?", mid-term memory answers "what am Related by memory
  2. Wiki concept Jack Roberts A YouTube creator and startup builder whose Codex and Claude Code walkthroughs frame AI tools as practical work systems: Codex as Related by memory
  3. Insight AI Measurement and Experimentation How to measure AI product impact with evals, adoption metrics, online experiments, guardrails, and cost tracking Readers have engaged with this next