Claude OS

AI Memory & Knowledge Base System for Claude Code
Initialize projects in seconds • Never lose context again

---

We have a Discord! Join our community to get help, share your projects, and connect with other Claude OS users.

---

🆕 What's New in v2.3

Latest Release: December 2025

🎯 Skills Library & Community Skills

Browse and install 36+ skills from the community with one click!

/claude-os-skills                    # List all skills
/claude-os-skills install pdf        # Install from community

New Features:

• 📚 Local Templates - Pre-built skills for Rails, React, testing workflows
• 🌐 Community Skills - Install from Anthropic Official (16) & Superpowers (20)
• 🔧 Custom Skills - Create and share project-specific skills
• 🎨 Beautiful UI - Browse, search, and install via web interface
• 📖 Recommended Skills - Our curated list of skills we actually use and trust

📊 Session Insights

Parse Claude Code sessions and extract insights automatically!

New Features:

• 🔍 Session Parser - Reads Claude Code's native .jsonl session files
• 💡 Insight Extraction - Extracts patterns, decisions, blockers
• 📈 Analytics Ready - Track tool usage and file changes across sessions

✨ Beautiful New Installer (v2.2)

The installer got a major upgrade with Charm CLI (gum) support!

./setup-claude-os.sh --demo    # See the beautiful UI without changes
./setup-claude-os.sh --dry-run # Preview what would happen
./setup-claude-os.sh --help    # See all options

New Features:

• 🎨 Gum Integration - Beautiful interactive menus with arrow-key navigation (with graceful bash fallback)
• 🛡️ Safety Features - --demo, --dry-run, and automatic config backups
• 💨 Lite Model Default - Now defaults to llama3.2:3b (2GB) instead of 8b (4.7GB)
• ☁️ Cloud Option - Choose between Local (Ollama) or Cloud (OpenAI) during setup
• 🐧 Better Linux Support - Improved package manager detection and installation

Safety First:

• --demo - Try the installer UI without making any changes
• --dry-run - See exactly what would be done before doing it
• Auto-backup - Existing .env files backed up before overwriting

📋 Recent Improvements

Version	Highlights
v2.3	Skills library, community skills, session insights
v2.2	Gum CLI support, safety features, lite model default
v2.1	Unified installer, OpenAI provider support
v2.0	Hybrid tree-sitter indexing, real-time Kanban board

🙌 Community Contributions

Thanks to our amazing contributors!

PR	Contributor	Description
#22	@illAssad	Fix delete document endpoint SQLite cursor handling
#21	@illAssad	Skip node_modules and build directories during ingestion
#20	@illAssad	Fix tree-sitter version compatibility
#19	@illAssad	Add non-blocking semantic indexing with Jobs Dashboard UI
#18	@illAssad	Fix frontend startup by auto-installing npm dependencies
#17	@williamclavier	Fix: Ensure commands/skills directories exist
#16	@jplimack	Fix hardcoded paths - make dynamic
#12	@gkastanis	Add missing frontend lib files
#11	@gkastanis	Add Linux support for installation
#10	@nicseltzer	Fix broken README link

---

🚀 What is Claude OS?

Claude OS is Claude Code's personal memory system - making AI the best coding assistant in the universe by remembering everything across sessions.

The Problem

You work with Claude Code on a feature, close the terminal, come back tomorrow... and Claude forgot everything. You explain the same architecture. You reference the same files. You repeat yourself constantly.

The Solution

Claude OS gives Claude persistent memory:

• 📝 Remembers decisions across all sessions
• 🔍 Searches past work automatically at session start
• 📚 Indexes your docs and makes them searchable
• 🧠 Learns patterns that improve over time
• 🔄 100% Local - Never leaves your machine, fully private

Key Features

✅ Skills Library - NEW! Browse & install skills from Anthropic and community repos ✅ Session Insights - NEW! Parse Claude Code sessions and extract patterns automatically ✅ Lightning-Fast Indexing - Tree-sitter hybrid indexing: 10,000 files in 30 seconds ✅ Real-Time Kanban Board - Auto-syncing task visualization for agent-os specs ✅ One-Command Project Init - /claude-os-init and you're done ✅ Automatic Context Loading - Starts every session with relevant memories ✅ Session Management - Track work, save progress, resume later ✅ Documentation Ingestion - Auto-indexes your docs during setup ✅ Agent-OS Integration - Optional spec-driven development with 8 specialized agents ✅ Flexible AI - Choose local Ollama (free) or OpenAI API (paid) ✅ Template System - Commands and skills shared via symlinks

---

⚡ NEW: Hybrid Indexing System

Claude OS v2.0 introduces lightning-fast tree-sitter based indexing!

The Problem with Traditional Indexing

Previous versions embedded EVERY file, which was painfully slow for large codebases:

• Large projects (10,000+ files): 3-5 hours to index
• Must complete before Claude can start working
• High resource usage, blocks productive coding

The Solution: Hybrid Two-Phase Indexing

Inspired by Aider's approach, Claude OS now uses:

Phase 1: Structural Index (30 seconds)

• ⚡ Parse files with tree-sitter (no LLM calls!)
• 📊 Extract symbols only (classes, functions, signatures)
• 🔗 Build dependency graph
• 🏆 PageRank importance scoring
• ✅ Ready to code immediately!

Phase 2: Semantic Index (optional, background)

• 🎯 Selective embedding (top 20% most important files)
• 📚 Full embedding for documentation
• 🔍 Deep semantic search when needed
• ⏰ Runs in background while you code

Performance Comparison

Feature	Before	After (Hybrid)
Large project (10k files)	3-5 hours	30 seconds + 20 min optional
Files embedded	100,000+ chunks	~20,000 chunks (80% reduction)
Start coding	After full index	Immediately!
Resource usage	High Ollama load	Minimal CPU/memory
Query speed	Semantic search	Instant structural + semantic

📖 Read the full design: docs/HYBRID_INDEXING_DESIGN.md

---

🎨 Visual Interface

Claude OS provides a beautiful, intuitive web interface for managing your AI development workflow:

Welcome Screen Get started with Claude OS	Project Overview View project details and MCP status
Kanban Board Track spec implementation progress	Services Dashboard Monitor all Claude OS services

📖 See the complete visual guide: docs/guides/VISUAL_GUIDE.md

---

🏗️ Architecture Overview

Claude OS is built on 5 core pillars that work together to give Claude persistent memory:

1. 🧠 Real-Time Learning - Automatically captures insights from conversations via Redis Pub/Sub
2. 💾 Memory MCP - Persistent memory system with instant recall using natural language
3. 🔍 Analyze-Project - Intelligent codebase indexing with git hooks and tree-sitter
4. 🎯 Session Management - Auto-resume sessions with full context preservation
5. 📚 Semantic Search - Vector-based code understanding and pattern recognition

All knowledge flows through the Semantic Knowledge Base (SQLite + sqlite-vec), exposed via the MCP Server (port 8051) to Claude Code, giving you an AI assistant that never forgets.

Data Flow: Git Commit → 3s indexing → SQLite → MCP → Claude → You

---

💻 Installation & Setup

Prerequisites

Required:

• macOS or Linux (Ubuntu, Debian, Fedora, RHEL, Arch)
• Python 3.11 or 3.12 (python3 --version)
  • Note: Python 3.13+ not yet supported due to dependency constraints
• Git (git --version)

Optional:

• Node.js 16+ (for React UI)
• Ollama (for local AI) or OpenAI API key

Note: Windows support coming soon.

Quick Installation

# Clone the repository
git clone https://github.com/brobertsaz/claude-os.git
cd claude-os

# Run the unified installer
./setup-claude-os.sh

First time? Try the demo first:

./setup-claude-os.sh --demo    # See the beautiful UI (no changes made)
./setup-claude-os.sh --dry-run # Preview what would be installed

The installer will guide you through:

1. Choose Provider: Local (Ollama) or Cloud (OpenAI)
2. Choose Model Size: Lite (2GB) or Full (4.7GB) - for local installs
3. Automatic Setup: Python, dependencies, MCP server, commands/skills

What gets installed:

• ✅ Python virtual environment
• ✅ All dependencies
• ✅ MCP server configuration
• ✅ Commands and skills symlinked to ~/.claude/
• ✅ Ollama + models (if local provider selected)
• ✅ Redis for caching

Installer Options

./setup-claude-os.sh           # Interactive installation
./setup-claude-os.sh --demo    # Try the UI without changes
./setup-claude-os.sh --dry-run # Preview what would happen
./setup-claude-os.sh --help    # Show all options
./setup-claude-os.sh --version # Show version

Legacy Scripts

The old install.sh and setup.sh scripts still work - they redirect to the new unified installer.

Visit http://localhost:5173 to use the web UI.

Starting Claude OS

After installation, start the services:

./start.sh

This starts the MCP server at http://localhost:8051

---

🚀 Quick Start

Initialize any project with Claude OS in under 2 minutes:

Step 1: Navigate to Your Project

cd /path/to/your/project

Step 2: Initialize with Claude OS

In Claude Code, run:

/claude-os-init

The command will:

1. Ask Questions Interactively:

   • Project name (auto-detects from folder)
   • Tech stack (Ruby on Rails, Python, Node.js, etc.)
   • Database (PostgreSQL, MySQL, etc.)
   • Development environment (Docker, Local, etc.)
   • Brief description
   • Documentation directory to ingest (optional)

2. Create Project in Claude OS:

   • Calls API to create project
   • Creates 4 knowledge bases automatically:
     • {project}-project_memories - Claude's memory
     • {project}-project_profile - Architecture & standards
     • {project}-project_index - Codebase index
     • {project}-knowledge_docs - Your documentation

3. Set Up Project Structure:

   your-project/
   ├── CLAUDE.md           # Auto-loaded every session!
   ├── .claude/            # Commands, skills, agents
   │   ├── ARCHITECTURE.md
   │   ├── CODING_STANDARDS.md
   │   └── DEVELOPMENT_PRACTICES.md
   └── .claude-os/         # Config and state (git-ignored)
       ├── config.json
       └── hooks.json
   

4. Ingest Documentation:

   • Scans your docs directory
   • Uploads all files to {project}-knowledge_docs
   • Creates vector embeddings for search

5. Analyze Codebase:

   • Runs initialize-project skill
   • Generates coding standards
   • Documents architecture
   • Indexes key files

6. Ready to Code:

   • Claude now knows your project
   • Memory persists across sessions
   • Context auto-loads on session start

What You Get

• ✅ 4 knowledge bases created (memories, profile, index, docs)
• ✅ Documentation auto-indexed
• ✅ Codebase analyzed
• ✅ CLAUDE.md file with all context
• ✅ Ready to code with AI memory!

---

🧠 How Claude OS Works

Session Workflow

Every Claude Code session automatically:

1. Checks for Active Session

   • Reads claude-os-state.json
   • Prompts: Continue working? Start something new?

2. Loads Context

   • Searches {project}-project_memories for recent work
   • Loads relevant patterns and decisions
   • Shows what it remembers

3. Works With Memory

   • Saves insights with /claude-os-remember
   • Searches memories with /claude-os-search
   • References past decisions automatically

4. Ends Session

   • Saves session summary
   • Updates memories
   • Tracks what was accomplished

Available Commands

All these work in any initialized project:

• /claude-os-init - Initialize new project
• /claude-os-search [query] - Search memories & docs
• /claude-os-remember [content] - Quick save to memories
• /claude-os-save [title] - Full-featured save with KB selection
• /claude-os-list - List all knowledge bases
• /claude-os-session [action] - Manage development sessions
• /claude-os-triggers - Manage trigger phrases
• /claude-os-skills [action] - Manage skills (list, install, create)

Available Skills

Global Skills (always available):

• initialize-project - Analyze codebase and generate standards
• remember-this - Auto-save when you say "remember this:"
• memory - Simple memory management

Community Skills (install via /claude-os-skills):

• 36+ skills from Anthropic Official and Superpowers repos
• PDF/XLSX manipulation, frontend design, TDD, debugging, code review, and more

---

🤖 Agent-OS: Spec-Driven Development (Optional)

Created by Builder Methods (CasJam Media LLC) MIT Licensed • Separate Optional Integration

Agent-OS adds structured workflows for planning and implementing features using 8 specialized agents.

Agent-OS is a separate open-source project that can be installed alongside Claude OS. We're grateful to Builder Methods for creating such powerful spec-driven development tools.

Manual Installation

If the Agent-OS repository is available, you can install it with:

git clone https://github.com/buildermethods/agent-os.git ~/.claude/agents/agent-os

Note: Check if the repository exists before attempting to install.

When to Use Agent-OS

If you have Agent-OS installed, use it when you want:

• Structured feature planning with iterative requirements gathering
• Detailed specifications before coding
• Task breakdowns with clear implementation steps
• Verification workflows to ensure completeness

The 8 Agents

Specification Workflow:

1. spec-initializer - Initialize new spec directories
2. spec-shaper - Gather requirements through 1-3 questions at a time
3. spec-writer - Create detailed technical specifications
4. tasks-list-creator - Break specs into actionable tasks

Implementation Workflow:

5. implementer - Implement features following task list
6. implementation-verifier - Verify implementation completeness
7. spec-verifier - Verify specs and tasks consistency
8. product-planner - Create product documentation

Agent-OS Commands

Available when enabled:

• /new-spec - Initialize a new feature specification
• /create-spec - Full specification workflow (gather requirements → create spec → generate tasks)
• /plan-product - Create product mission, roadmap, and tech stack docs
• /implement-spec - Implement a specification following its tasks

How It Works

1. User: "/new-spec user-authentication"
   → Agent creates spec directory structure

2. User: "/create-spec"
   → spec-shaper asks 1-3 questions at a time
   → Gathers requirements iteratively
   → Identifies reusable code
   → Collects visual assets

3. Agent: spec-writer creates detailed specification
   → tasks-list-creator generates actionable tasks

4. User: "/implement-spec user-authentication"
   → implementer follows tasks step-by-step
   → implementation-verifier checks completeness

5. Result: Fully specified, implemented, and verified feature!

Agent-OS Project Structure

When enabled, your project gets:

your-project/
├── agent-os/
│   ├── config.yml          # Agent-OS configuration
│   ├── product/            # Product documentation
│   │   ├── mission.md      # Product mission
│   │   ├── roadmap.md      # Feature roadmap
│   │   └── tech-stack.md   # Technology stack
│   ├── specs/              # Feature specifications
│   │   └── YYYY-MM-DD-feature-name/
│   │       ├── planning/
│   │       │   ├── requirements.md
│   │       │   └── visuals/
│   │       ├── spec.md
│   │       └── tasks.md
│   └── standards/          # Coding standards (as skills)
└── .claude/agents/agent-os/  # 8 agents (symlinked)

Integration with Claude OS

Agent-OS agents deeply integrate with Claude OS:

• Search memories before creating specs (avoid reinventing)
• Save decisions to project_memories during planning
• Reference patterns from previous work
• Build knowledge that improves over time

This is the complete AI development system!

---

🎯 Skills Library

Browse, install, and manage Claude Code skills with ease!

What Are Skills?

Skills are reusable instruction sets that teach Claude specific capabilities. They can include:

• Coding patterns and best practices
• Tool usage workflows
• Domain-specific knowledge
• Development methodologies

Skill Types

Global Skills (~/.claude/skills/)

• Available in ALL projects
• Core skills: memory, remember-this, initialize-project

Project Skills ({project}/.claude/skills/)

• Available only in that project
• Installed from templates or custom created

Community Skills (fetched from GitHub)

• Anthropic Official - 16 skills from anthropics/skills
• Superpowers - 20 skills from obra/superpowers

Using the Skills Command

# List all installed skills
/claude-os-skills

# Browse local templates
/claude-os-skills templates

# Install a template to your project
/claude-os-skills install rails-backend

# Create a custom skill
/claude-os-skills create

# View skill details
/claude-os-skills view <name>

# Delete a project skill
/claude-os-skills delete <name>

Community Skills (via Web UI)

1. Open the web UI at http://localhost:5173
2. Select your project
3. Click the Skills tab
4. Click Install Template
5. Switch to Community Skills tab
6. Browse skills from Anthropic Official and Superpowers
7. Click Install on any skill

Featured Community Skills

From Anthropic Official:

• pdf - Create, edit, and analyze PDF documents
• xlsx - Spreadsheet manipulation with formulas
• frontend-design - Production-grade UI components
• mcp-builder - Create MCP servers
• doc-coauthoring - Collaborative documentation

From Superpowers:

• test-driven-development - TDD workflow
• systematic-debugging - Four-phase debugging framework
• code-review - Rigorous code review process
• git-worktrees - Isolated development branches
• brainstorming - Structured ideation process

---

🎯 Spec Tracking & Kanban Board

NEW: Real-time auto-syncing Kanban board for Agent-OS specs!

Visual Kanban board showing specs, tasks, and progress tracking

What It Does

When you use Agent-OS to create specs with /create-spec, Claude OS automatically:

• 📋 Parses tasks.md files - Extracts all tasks, phases, dependencies, and metadata
• 🗄️ Stores in database - Tracks progress, completion, and time estimates
• 📊 Displays as Kanban - Visual board showing specs and tasks by status
• ⚡ Real-time sync - NEW! Auto-detects file changes and updates within 3 seconds
• 👀 File watching - Monitors agent-os/specs/ folder for changes
• ✅ Auto-refresh - Board polls every 3 seconds for live updates
• 🗃️ Archives completed specs - Keep your board focused on active work

Features

Real-Time File Watching (NEW!):

• Automatically monitors your agent-os/specs/ folder
• Detects changes to tasks.md and spec.md files
• 2-second debounce to batch rapid edits
• Auto-syncs to database within 3 seconds
• Frontend auto-refreshes every 3 seconds
• Total latency: ~6 seconds from file save to board update

Automatic Syncing:

• Syncs all specs from your project's agent-os/specs/ folder
• Tracks task metadata (estimated time, dependencies, risk level)
• Auto-detects completed tasks (marked with ✅ or [x] in tasks.md)
• Supports both checkbox format and classic format

Progress Tracking:

• Status auto-updates based on completion:
  • planning - No tasks completed yet
  • in_progress - Some tasks completed
  • completed - All tasks done
• Progress percentage calculated automatically
• Time estimates tracked (estimated vs actual minutes)

Archive Feature:

• Archive completed specs to keep your board clean
• Archived specs hidden by default but can be viewed
• Preserves all task history for future reference

API Endpoints

All spec tracking functionality is exposed via REST API:

# Get all specs for a project
GET /api/projects/{project_id}/specs

# Get all tasks for a spec
GET /api/specs/{spec_id}/tasks

# Update task status
PATCH /api/tasks/{task_id}/status
{
  "status": "in_progress",  # todo, in_progress, done, blocked
  "actual_minutes": 15
}

# Sync specs from agent-os folder (manual)
POST /api/projects/{project_id}/specs/sync

# Get Kanban board view
GET /api/projects/{project_id}/kanban?include_archived=false

# Archive/unarchive specs
POST /api/specs/{spec_id}/archive
POST /api/specs/{spec_id}/unarchive

# NEW: Real-time spec watcher control
GET /api/spec-watcher/status
POST /api/spec-watcher/start/{project_id}
POST /api/spec-watcher/stop/{project_id}
POST /api/spec-watcher/start-all

See: docs/guides/REALTIME_KANBAN_GUIDE.md for complete documentation.

How It Works

1. You create a spec with Agent-OS:
   /create-spec → agent-os/specs/2025-01-15-user-auth/

2. Spec Watcher detects the new folder:
   - Auto-starts when MCP server boots
   - Monitors agent-os/specs/ directory
   - 2-second debounce for batch changes

3. Auto-sync to database:
   - Reads tasks.md
   - Parses checkbox format: - [x] Task title
   - Extracts metadata, tasks, phases
   - Stores in SQLite database
   - ✅ Completes within 3 seconds

4. View in Kanban board (auto-refreshes every 3 seconds):
   - Todo: PHASE1-TASK1, PHASE1-TASK2
   - In Progress: PHASE2-TASK1
   - Done: PHASE1-TASK3, PHASE1-TASK4

5. As you work, agent-os updates tasks.md:
   - File watcher detects change
   - Auto-syncs to database
   - Board refreshes automatically
   - Total latency: ~6 seconds

6. Archive when complete:
   - Mark spec as archived
   - Keeps history but cleans up board

Database Schema

Two new tables track specs and tasks:

specs table:

• Stores spec metadata (name, path, status)
• Tracks total/completed tasks
• Calculates progress percentage
• Archive flag to hide completed specs

spec_tasks table:

• Individual tasks with codes (PHASE1-TASK1)
• Status (todo/in_progress/done/blocked)
• Time tracking (estimated vs actual)
• Dependencies between tasks
• Risk levels and phases

Example Usage

# Sync all specs for your project
curl -X POST http://localhost:8051/api/projects/1/specs/sync

# Response:
{
  "synced": 3,
  "updated": 0,
  "total": 3,
  "errors": []
}

# Get Kanban view
curl http://localhost:8051/api/projects/1/kanban

# Response shows:
# - Your specs with tasks
# - Tasks grouped by status
# - Progress percentages
# - Time estimates

This is the complete AI development system!

---

📂 Template System

claude-os/
├── templates/              # Shared templates
│   ├── commands/          # Slash commands (symlinked to ~/.claude/)
│   │   ├── claude-os-init.md
│   │   ├── claude-os-search.md
│   │   ├── claude-os-skills.md    # NEW: Skills management
│   │   └── ...
│   ├── skills/            # Global skills (symlinked to ~/.claude/)
│   │   ├── initialize-project/
│   │   ├── remember-this/
│   │   └── memory/
│   ├── skill-library/     # NEW: Local skill templates
│   │   ├── general/       # General purpose skills
│   │   ├── rails/         # Ruby on Rails skills
│   │   ├── react/         # React/TypeScript skills
│   │   └── testing/       # Testing frameworks
│   └── project-files/     # Files created during /claude-os-init
│       ├── CLAUDE.md.template
│       └── .claude-os/
│           ├── config.json.template
│           └── hooks.json.template
├── cli/                   # CLI tools
│   └── claude-os-consolidate.sh
├── install.sh             # Quick setup script
└── start.sh               # Start services

Benefits:

• ✅ Update once, all projects benefit
• ✅ Symlinks mean instant updates
• ✅ Consistent across projects

---

📚 Managing Knowledge Bases

Via Web UI

1. Visit http://localhost:5173
2. Create Knowledge Base:
   • Click "Create Knowledge Base"
   • Choose type (Generic, Code, Documentation, Agent_OS)
3. Upload Documents:
   • Select KB from dropdown
   • Drag & drop files or click upload
   • Supports .md, .txt, .pdf, .py, .js, .ts, .json, .yaml
4. Query:
   • Type question in search box
   • View answer with source citations

Via CLI

# Search your project memories
/claude-os-search "how did we implement authentication?"

# Save a quick insight
/claude-os-remember "Fixed bug in user controller by adding validation"

# Full-featured save
/claude-os-save "Authentication Pattern" my-app-project_profile Architecture

Auto-Created KBs

When you run /claude-os-init, you get 4 knowledge bases:

1. {project}-project_memories

   • Claude's memory for decisions, patterns, solutions
   • Automatically saved during sessions
   • Searched at session start

2. {project}-project_profile

   • Architecture, coding standards, practices
   • Generated by initialize-project skill
   • Updated as project evolves

3. {project}-project_index

   • Automated codebase index
   • Tracks file structure
   • Updates on git commits (with hooks)

4. {project}-knowledge_docs

   • Your documentation
   • Auto-ingested during init
   • Add more via UI or CLI

---

⚙️ Configuration

Environment Variables

# Provider (local = Ollama, openai = OpenAI API)
CLAUDE_OS_PROVIDER=local            # Default: local

# SQLite Database
SQLITE_DB_PATH=data/claude-os.db    # Default: data/claude-os.db

# Ollama (for local provider)
OLLAMA_HOST=http://localhost:11434  # Default: localhost:11434
OLLAMA_MODEL=llama3.2:3b            # Default: llama3.2:3b (lite model)
OLLAMA_EMBED_MODEL=nomic-embed-text # Default: nomic-embed-text

# OpenAI (for openai provider)
OPENAI_API_KEY=sk-...               # Required if using OpenAI
OPENAI_LLM_MODEL=gpt-4o-mini        # Default: gpt-4o-mini
OPENAI_EMBED_MODEL=text-embedding-3-small  # Default

# MCP Server
MCP_SERVER_HOST=0.0.0.0             # Default: 0.0.0.0
MCP_SERVER_PORT=8051                # Default: 8051

Project Configuration

Each project has .claude-os/config.json:

{
  "project_name": "my-app",
  "claude_os_url": "http://localhost:8051",
  "knowledge_bases": {
    "memories": "my-app-project_memories",
    "profile": "my-app-project_profile",
    "index": "my-app-project_index",
    "docs": "my-app-knowledge_docs"
  },
  "docs_settings": {
    "watch_paths": ["./docs", "./knowledge_docs"],
    "auto_ingest_patterns": ["*.md", "*.txt", "*.pdf"]
  },
  "tech_stack": "Ruby on Rails",
  "database": "MySQL"
}

---

📊 Performance

Native Ollama Setup:

• Response time: ~40 seconds per query
• GPU acceleration: Full Metal GPU on Apple Silicon
• Memory usage: 8-10GB (models + context)
• CPU usage: 12 cores (M4 Pro)

Why it's fast:

• Direct GPU acceleration (no virtualization)
• Efficient vector search in SQLite
• Optimized RAG engine with caching
• Single-file database with minimal overhead

---

🛠️ Scripts Guide

Installation & Setup

./setup-claude-os.sh - Unified Installer (Recommended)

./setup-claude-os.sh           # Interactive installation
./setup-claude-os.sh --demo    # Try the beautiful UI (no changes)
./setup-claude-os.sh --dry-run # Preview what would happen
./setup-claude-os.sh --help    # Show all options
./setup-claude-os.sh --version # Show version (v2.2.0)

Features:

• 🎨 Beautiful interactive UI with Charm CLI (gum) support
• 🛡️ Safety features: --demo, --dry-run, automatic config backups
• ☁️ Provider choice: Local (Ollama) or Cloud (OpenAI)
• 💨 Model choice: Lite (llama3.2:3b, 2GB) or Full (llama3.1:8b, 4.7GB)
• 🐧 Cross-platform: macOS and Linux (Ubuntu, Debian, Fedora, RHEL, Arch)

What it installs:

• ✅ Python virtual environment + dependencies
• ✅ Ollama + AI models (if local provider)
• ✅ Redis for caching
• ✅ MCP server configuration
• ✅ Commands and skills symlinked to ~/.claude/

Legacy Scripts

The old scripts redirect to the unified installer:

./install.sh  # → redirects to setup-claude-os.sh
./setup.sh    # → redirects to setup-claude-os.sh

Service Management

./start.sh or ./start_all_services.sh - Start Everything

./start.sh

Starts:

• 🔌 MCP Server (port 8051)
• 🎨 React Frontend (port 5173)
• 🤖 RQ Workers
• 💾 Redis
• 🧠 Ollama

./stop_all_services.sh - Stop All

./stop_all_services.sh

./restart_services.sh - Restart

./restart_services.sh

---

🗑️ Uninstalling Claude OS

To completely remove Claude OS from your system:

cd /path/to/claude-os
./uninstall.sh

The uninstall script removes:

• Command symlinks from ~/.claude/commands/
• Skill symlinks from ~/.claude/skills/
• MCP server config from ~/.claude/mcp-servers/
• Python virtual environment (venv/)
• Config files and logs
• Optionally: your knowledge base data

What it does NOT remove:

• The claude-os/ directory itself (delete manually with rm -rf)
• Ollama (see Ollama uninstall docs)
• Redis (brew uninstall redis on macOS)

Manual Uninstall:

If you prefer to uninstall manually:

# Remove symlinks
rm ~/.claude/commands/claude-os-*.md
rm -rf ~/.claude/skills/initialize-project
rm -rf ~/.claude/skills/remember-this
rm -rf ~/.claude/skills/memory
rm ~/.claude/mcp-servers/code-forge.json

# Remove Claude OS directory
rm -rf /path/to/claude-os

---

🐛 Troubleshooting

"Command not found: /claude-os-init"

Symlinks weren't created. Re-run:

cd /path/to/claude-os
./install.sh

"Connection refused to localhost:8051"

Claude OS server isn't running:

cd /path/to/claude-os
./start.sh

"Project already exists"

Project name is taken. Choose a different name or delete via UI at http://localhost:5173

Port Already in Use

# Find process on port 8051
lsof -i :8051

# Kill if needed
kill -9 <PID>

Ollama Issues

# Check if running
ollama list

# Start manually
ollama serve

# Check for model
ollama list | grep llama3.1

---

📁 Project Structure

claude-os/
├── templates/              # Shared templates system
│   ├── commands/          # Slash commands
│   ├── skills/            # Global skills
│   ├── skill-library/     # Local skill templates (NEW)
│   └── project-files/     # Files created during init
├── cli/                   # CLI tools
│   └── claude-os-consolidate.sh
├── app/                    # Backend application
│   ├── core/              # Core modules
│   │   ├── sqlite_manager.py
│   │   ├── rag_engine.py
│   │   ├── skill_manager.py    # NEW: Skills management
│   │   ├── session_parser.py   # NEW: Session parsing
│   │   ├── insight_extractor.py # NEW: Insight extraction
│   │   └── ...
│   └── db/                # Database schemas
├── frontend/              # React UI (Vite)
│   ├── src/
│   │   ├── components/
│   │   │   ├── SkillsManagement.tsx  # NEW: Skills UI
│   │   │   └── ...
│   │   └── pages/
│   └── public/
│       └── assets/
├── mcp_server/           # MCP Server (HTTP)
│   └── server.py         # FastAPI + MCP endpoints
├── data/                 # SQLite database
│   └── claude-os.db
├── logs/                 # Service logs
├── install.sh            # Quick setup script
├── start.sh              # Start services
└── README.md             # This file

---

📖 Additional Documentation

Getting Started

• templates/README.md - 📂 Template system documentation

Core Features

• docs/guides/REALTIME_KANBAN_GUIDE.md - ⚡ NEW! Real-time Kanban board (auto-sync, file watching, API reference)
• docs/SELF_LEARNING_SYSTEM.md - 🧠 How Claude learns automatically
• docs/REAL_TIME_LEARNING_GUIDE.md - Real-time learning usage

Technical Documentation

• docs/API_REFERENCE.md - 🔌 Complete API Reference (all endpoints, examples, authentication)
• docs/HYBRID_INDEXING_DESIGN.md - ⚡ Hybrid indexing architecture
• README_NATIVE_SETUP.md - Detailed native setup
• NATIVE_VS_DOCKER_DECISION.md - Why native Ollama
• PERFORMANCE_TEST_RESULTS.md - Benchmark results

---

🤝 Contributing

Claude OS is open source. Feel free to:

• Modify for your specific needs
• Add new commands and skills
• Optimize RAG strategies
• Contribute improvements back

---

🙏 Acknowledgments

Agent-OS Integration

Claude OS optionally integrates with Agent-OS by Builder Methods (CasJam Media LLC).

• Project: Agent-OS - Spec-driven development workflow system
• Author: Builder Methods (CasJam Media LLC)
• License: MIT
• Repository: https://github.com/builder-methods/agent-os

Agent-OS provides 8 specialized agents for structured feature planning and implementation. We're grateful to Builder Methods for creating such powerful tools and for licensing them under MIT, making this integration possible.

If you find Agent-OS valuable, consider:

• ⭐ Starring their repository
• 📣 Sharing it with other developers
• 🤝 Contributing to their project

---

📄 License

MIT License - Use it freely!

Note: This project (Claude OS) is MIT licensed. Agent-OS, when installed, is a separate project also MIT licensed by Builder Methods (CasJam Media LLC). See the Agent-OS repository for their specific license terms.

---

Claude Code + Claude OS = Invincible! 🚀
Built by AI coders, for AI coders