Ogmios - Personal AI Infrastructure

Transform Claude Code into a specialized AI system with persistent context, autonomous agents, and automation.

Features

• Specialized Skills: 16+ domain experts (engineering, research, security, etc.)
• Autonomous Agents: 7 research agents that work independently
• Slash Commands: Quick actions via /refactor, /security-review, etc.
• Memory System: Persistent decisions, learnings, and preferences
• Automation: Event-driven hooks
• Voice System: Multi-voice feedback (optional)
• Bilingual: English and Swedish support

Quick Start

Prerequisites

• Claude Code installed
• Bun for TypeScript hooks
• macOS or Linux

Installation

# Clone repository
git clone https://github.com/carlheath/ogmios.git
cd ogmios

# Install to home directory
cp -r .claude/ ~/.claude/

# Install hook dependencies
cd ~/.claude/hooks
bun install

# Customize your AI identity
nano ~/.claude/skills/CORE/SKILL.md  # Replace <PLACEHOLDERS>

# Start Claude Code
claude

Core Systems

Skills = Domain Expertise

Specialized AI personas with deep knowledge:

Skill	Purpose	Triggers
engineering	Code implementation	"build", "implement", "fix"
architecture	System design	"design", "architect"
security	Security audits	"security", "vulnerability"
academic	Academic writing	"thesis", "literature review"
data-analysis	Data science	"analyze", "visualize"
document-factory	Create PRDs, RFCs, ADRs	"create document"

Agents = Autonomous Workers

Research agents that gather information independently:

Agent	Purpose
code-engineer	Code review and quality
security-engineer	Security analysis
academic-researcher	Literature reviews
ml-researcher	ML/AI model research
person-researcher	Professional background research
ux-researcher	Design patterns research
test-engineer	Test coverage analysis

Commands = Quick Actions

Slash commands for common tasks:

Command	Purpose
/refactor	Refactor code with best practices
/security-review	OWASP-focused security audit
/update-check	Check for Claude Code updates
/web-search	Structured web research

Memory = Persistent Context

~/.claude/memory/
├── decisions/    # Architectural decisions (ADRs)
├── learnings/    # Patterns and mistakes
├── preferences/  # Tech stack, style
└── procedures/   # Recurring workflows

Hooks = Automation

Event-driven scripts:

• SessionStart - Load context automatically
• UserPromptSubmit - Activate relevant skills
• Stop - Cleanup and notifications

Directory Structure

~/.claude/
├── skills/           # Domain expertise (16+)
│   ├── CORE/         # Your AI's identity
│   ├── engineering/  # Code skills
│   ├── security/     # Security expertise
│   └── ...
├── agents/           # Autonomous workers (7)
├── commands/         # Slash commands (4)
├── hooks/            # Automation scripts
├── memory/           # Persistent context
│   ├── decisions/
│   ├── learnings/
│   ├── preferences/
│   └── procedures/
└── documentation/    # System docs

Customization

1. Define Your AI Identity

Edit ~/.claude/skills/CORE/SKILL.md:

Your Name: <AI_ASSISTANT_NAME>
Your Role: <AI_ROLE_DESCRIPTION>
Personality: <PERSONALITY_TRAITS>

2. Set Tech Preferences

Create ~/.claude/memory/preferences/stack.md:

## Languages
- Primary: TypeScript (bun)
- Secondary: Python (uv)

3. Record Decisions

Create ~/.claude/memory/decisions/my-decision.md:

# Decision: Use TypeScript for CLI

## Context
Need consistent tooling.

## Decision
TypeScript with Bun.

## Rationale
Fast, typed, npm ecosystem.

Philosophy

Built on these principles:

1. System > Model - Architecture matters more than raw AI
2. Skills Before Search - Load expertise, don't search for "how to"
3. Delegate Research - Agents gather, main context synthesizes
4. Code Before Prompts - Write code to solve, prompts to orchestrate

Documentation

• Quick Start
• Installation
• System Architecture
• Troubleshooting

Security

• Local-first design (data stays on your machine)
• No API keys in repository
• See SECURITY.md for vulnerability reporting

Credits

Inspired by: Daniel Miessler's Kai

Thanks to:

• Daniel Miessler for Kai architecture
• Anthropic for Claude and Claude Code
• MCP community for protocol development

License

MIT - see LICENSE