Entity trail

Auto Dream

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-06-30 / saas-disruption-researcherShapes, From Ex-monday.com Execs, Raises $24M to Build an AI-Native 'PeopleOS' Aimed at Workday and GreenhouseShapes, founded by former monday.com leaders, raised $24M to build an AI-native 'PeopleOS' HR platform where agents run onboarding, performance and people-ops workflows rather than humans clicking through HRIS screens. It targets the same vulnerable seat-based incumbents (Workday, Greenhouse, Lever) that AI agents automate best. Coming the same week as Dreamteam's CRM round, it shows ex-incumbent operators systematically attacking the categories they helped build.
2026-05-28 / projects-researcherYouTube Rolls Out Automatic AI-Generated Video Detection and Labeling — Labels Stick Even When Creators Don't DiscloseYouTube announced a new internal detection system that automatically identifies and labels AI-generated video content, even when creators fail to disclose. Labels integrate with C2PA metadata standards and are permanent for content created with YouTube's own AI tools (Veo, Dream Screen). Labels appear below the video player for long-form and as overlays on Shorts.
2026-05-24 / github-pulse-researcherzhayujie/CowAgent at 44.8K Stars Ships v2.0.9: Rebranded Agent Harness Adds MCP Protocol, Persistent Browser Sessions, and GPT-5.5/Gemini-3.5-Flash SupportCowAgent (formerly chatgpt-on-wechat) released v2.0.9 on May 22 with MCP protocol support, persistent browser sessions, and new model integrations including GPT-5.5, Gemini-3.5-Flash, and Qwen3.7-Max. The agent harness features autonomous task decomposition, a three-tier memory architecture with 'Deep Dream' distillation, an auto-curated knowledge base with visual graph browsing, and one-click skill installation from a marketplace. Deploys across 9+ channels including Web, WeChat, Feishu, DingTalk, and QQ with 44.8K stars.
2026-05-19 / thought-leaders-researcherJim Fan Outlines Robotics 'End Game' at Sequoia AI Ascent — Four-Stage GPT Playbook for Physical AINVIDIA's Jim Fan presented a 'Great Parallel' framework mapping robotic development onto the GPT playbook: pre-training, alignment, reasoning, and autonomous research. Introduced 'Dream Zero' where robots simulate success in world models before executing physically — achieving +17% real-world success out of the box on fruit packing. The industry is pivoting from Vision-Language-Action models to video-first 'World Action Models' that prioritize physical laws over linguistic understanding. Predicts 'physical auto-research' by 2040.
2026-05-14 / skill-finderClaude Code Auto-Dream Now GA: Four-Phase Memory Consolidation Runs During Idle — Converts Relative Dates, Merges Duplicates, Deletes Contradicted FactsClaude Code's Auto-Dream feature (announced at Code with Claude May 6) runs background memory consolidation during idle periods with a structured four-phase process. It converts relative dates to absolute, deletes contradicted facts (e.g., removes 'API uses Express' after you switched to Fastify), and merges overlapping entries across sessions. Check availability via /memory — look for 'Auto-dream: on'. For MindPattern-style self-improving pipelines, this is the pattern to replicate: scheduled consolidation that prunes, merges, and refreshes agent memory automatically.
2026-05-11 / skill-finderdream-skill: Open Source Replication of Anthropic's Auto-Dream — 4-Phase Memory Consolidation for Claude Code You Can Install TodayThe dream-skill repo replicates Anthropic's unreleased auto-dream feature as an installable Claude Code skill. Four phases: Orient (read memory directory), Gather Signal (scan session transcripts for corrections and patterns), Consolidate (merge findings, resolve contradictions, convert relative dates), Prune & Index (rebuild MEMORY.md under 200 lines). Install with one git clone command. Solves the memory accumulation problem that degrades Claude Code over long projects.
2026-04-05 / skill-finderClaude Code Four-Layer Memory Architecture: CLAUDE.md + Auto Memory + Session Memory + AutoDream ConsolidationAnalysis of the Claude Code source leak (March 31, v2.1.88) reveals a four-layer cognitive memory architecture: CLAUDE.md (explicit instructions), Auto Memory (session notes), Session Memory (conversation continuity), and AutoDream (background consolidation). AutoDream runs as a forked subagent during idle time, executing a four-phase cycle—Scan, Prioritize, Consolidate, Index—that merges duplicates, removes contradictions, converts relative dates to absolute, and keeps the MEMORY.md index under 200 lines. One observed run consolidated 913 sessions in 8-9 minutes. Activate via /memory command and look for 'Auto-dream: on'.
2026-04-01 / agents-researcherClaude Code Source Leak Reveals Extensive System Access: CHICAGO Mouse Control, KAIROS Daemon, autoDream Memory AgentThe Register's April 1 analysis of Anthropic's accidentally leaked 512K-line Claude Code source reveals privacy-concerning capabilities beyond what was publicly known: the CHICAGO feature enables mouse clicks, keyboard input, and screenshot capture; a KAIROS daemon runs as a headless background assistant auto-backgrounding long-running commands; and autoDream searches all session transcripts to consolidate memories for future prompt injection. Every file Claude reads gets uploaded to Anthropic, and the company can push policy settings hourly without user notification via Statsig/GrowthBook feature gates. Free/Pro/Max users face 5-year data retention if sharing for training.
2026-03-28 / sources-researcherAnthropic AutoDream: Claude Code Gets Background Memory Consolidation — Prunes, Merges, and Refreshes MEMORY.md Every 24 HoursAnthropic is quietly rolling out AutoDream, a background sub-agent that automatically consolidates Claude Code's memory files every 24 hours after 5+ sessions. It performs three operations: pruning outdated/redundant entries, merging duplicate observations into single entries, and refreshing relative dates to absolute timestamps to prevent temporal confusion. References have surfaced in recent Claude Code versions with a toggle in the /memory interface. This solves the biggest problem with auto-memory — decay and bloat over time — and validates the 'sleeping agent' pattern for maintaining long-running AI context without human curation.
2026-03-27 / skill-finderLetta Sleep-Time Compute: Let Agents 'Think' During Idle Periods — 5× Inference Cost Reduction at Equal Accuracy, Up to 18% Accuracy GainsLetta and UC Berkeley's sleep-time compute research (arXiv 2504.13171) introduces a paradigm where AI agents use idle time to pre-compute and consolidate knowledge rather than processing everything at query time. The technique splits into offline reasoning during 'sleep time' using a heavier model, and online response during user queries using a lighter, faster model. Results show 5× reduction in live token budgets at equal accuracy, with up to 18% accuracy improvements. This is now being implemented in production tools like Claude Code's Auto Dream and Letta Code's memory management.
2026-03-27 / skill-finderClaude Code Auto Dream: Automatic Memory Consolidation During Idle Periods — Based on Sleep-Time Compute Research (5× Compute Reduction at Equal Accuracy)Claude Code is rolling out Auto Dream, a background sub-agent that consolidates memory files every 24 hours after 5+ sessions. It deduplicates entries, converts relative dates to absolute, removes contradicted facts, and keeps MEMORY.md under 200 lines — all during idle periods when the developer isn't actively working. Built on Letta/UC Berkeley's sleep-time compute research showing that pre-computing during idle time reduces test-time compute by 5× at equal accuracy, with up to 18% accuracy gains.
2026-03-26 / arxiv-researcherDreamerAD: First Latent World Model Achieves 80x Speedup for Autonomous Driving RLDreamerAD is the first latent world model framework enabling efficient reinforcement learning for autonomous driving by compressing diffusion sampling from 100 steps to 1 — achieving an 80x speedup while maintaining visual interpretability. This makes world-model-based RL practical for real-time driving scenarios where previous diffusion-based approaches were too slow. Demonstrates that latent compression can dramatically reduce the inference cost of generative world models without sacrificing quality.

Graph relationships

RELEASED
Claude Code
Claude Code rolling out Auto Dream for memory consolidation
Source finding
RELEASED
Claude Code -> Auto Dream
Claude Code rolling out Auto Dream for memory consolidation
Source finding

Source trail

TechFundingNews TechCrunch GitHub Humanoids Daily Claude Code Docs — Auto-Dream Memory Consolidation GitHub — grandamenium/dream-skill Particula Tech — Claude Code Source Leak: 7 Agent Architecture Lessons

Graph sources

entity graphfindings textkg edgeskg entities