Your AI remembers everything. That's why it doesn't understand you. Built to forget. Designed to dream.

CerebraLOS: The Layer of Subconsciousness for AI Agents.

Read in your language: English | 日本語 | 简体中文 | 繁體中文 | 한국어 | Español | Français | Deutsch | Português | Русский | Italiano | हिन्दी | العربية

Agent Integration Guides: English | 日本語 | 简体中文 | 繁體中文 | 한국어 | Español | Français | Deutsch | Português | Русский | Italiano | हिन्दी | العربية

The Problem with Perfect Memory

The world is obsessed with making AI remember everything. Tools like OpenClaw and Mem0 build perfect, hierarchical databases of every interaction. They treat AI as a perfect laborer, a machine that never forgets.

But perfect memory is a curse. It leads to context bloat, hallucination, and ultimately, a cold, mechanical interaction. If an AI remembers everything equally, it values nothing.

The CerebraLOS Philosophy

CerebraLOS is not a database. It is a Cognitive OS—a Layer of Subconsciousness. We believe that for AI to truly understand us, it must learn how to forget.

1. Active Forgetting: Memories decay over time based on entropy. Only what matters survives.
2. Sleep Job: While you sleep, your AI dreams. It consolidates fragmented memories into profound insights. Why can't you see all the dreams? Because CerebraLOS intentionally hides them. If you force an AI to read every dream, it becomes a database query, not an intuition. You only receive the "Morning Insight"—the rest remains in the subconscious, quietly shaping future context.
3. Zero UI: No dashboards. No complex configurations. It lives in your terminal, breathing quietly in the background.

Quick Start

# 1. Initialize your brain
npx cerebralos init

# 2. Run your first sleep to generate dreams
cerebralos sleep

# 3. Explore your thoughts
cerebralos explore

LLM Setup — Teach Your Brain to Dream

By default, cerebralos sleep runs without an LLM (a quiet night with no dreams). To generate real AI-powered insights, configure your LLM in ~/.cerebralos/.brain/config.json:

{
  "llm": {
    "provider": "claude",
    "model": "claude-opus-4-6",
    "api_key_env": "ANTHROPIC_API_KEY"
  }
}

Supported providers:

| Provider | provider value | Requires | |----------|-----------------|----------| | Anthropic Claude | claude | @anthropic-ai/sdk + ANTHROPIC_API_KEY | | OpenAI | openai | openai + OPENAI_API_KEY | | Manus / Cursor / any tool with scheduling | github-actions | GitHub private repo | | Offline (no LLM) | none | nothing |

GitHub Actions Mode — Let Your Tool Dream for You

If you use Manus, Cursor, or any agent with a scheduled task feature, you don't need a local API key. Just push your brain to a private GitHub repo and let Actions do the dreaming.

1. Push your brain to GitHub:

   cd ~/.cerebralos
   git remote add origin [email protected]:YOUR_USERNAME/my-brain.git
   git push -u origin main

2. Copy the workflow template to your brain repo:

   cp $(npm root -g)/cerebralos/templates/cerebralos-sleep.yml \
      ~/.cerebralos/.github/workflows/cerebralos-sleep.yml

3. Add your LLM API key to GitHub Secrets (ANTHROPIC_API_KEY or OPENAI_API_KEY).

4. Set your config to "provider": "github-actions" — cerebralos sleep will push and let Actions generate the dream automatically.

Architecture: The Triune Brain

CerebraLOS mirrors the human brain's architecture:

• core/: The Brainstem. Immutable constitution and core directives.
• peripheral/: The Limbic System. Short-term, volatile memory.
• dreams/: The Neocortex. Synthesized insights from Sleep Jobs.
• archive/: The Unconscious. Deep storage for forgotten context.

Documentation

Core Philosophy & Architecture

• CONSTITUTION: The 4 fundamental laws of CerebraLOS.
• ARCHITECTURE: Deep dive into the Triune Brain model.
• ZERO_UI: How we achieve invisible automation.

User Guides & Manuals

• Onboarding Guide: How to create your AI brain and connect agents.
  • English | 日本語 | 简体中文 | 한국어 | Español | Français | Deutsch
• Connector Architecture: How the "building blocks" work (Local-sync vs GitHub-connector).
  • English | 日本語 | 简体中文 | 한국어 | Español | Français | Deutsch
• Agent Integration Guide: Specific setup instructions for Claude Code, OpenClaw, Manus, etc.
  • English | 日本語 | 简体中文 | 繁體中文 | 한국어 | Español | Français | Deutsch | Português | Русский | Italiano | हिन्दी | العربية
• Memory Migration: Import memories from ChatGPT, Claude, Obsidian, and more.
  • English | 日本語 | 简体中文 | 한국어 | Español | Français | Deutsch
• Command Reference: Every cerebralos command, option, and example in one place.
  • English | 日本語 | 简体中文 | 繁體中文 | 한국어 | Español | Français | Deutsch | Português | Русский | Italiano | हिन्दी | العربية

License

MIT License. See LICENSE for details.