Engram

The AI Harness for Everyone -- open-source personal AI infrastructure that turns any model into your AI.

Engram is the stable layer between a raw AI model and a useful AI system. It provides the structure -- skills, hooks, memory, security, personality -- so you can swap models without losing your setup. When a new model releases, your skills still work. Your memory persists. Your identity carries over.

Install

npm install -g engram-harness

Requires Node.js 20+.

Commands

engram init

Scaffolds your personal AI infrastructure at ~/.claude/. Creates identity files, starter skills, lifecycle hooks, memory directories, and a security validator -- everything needed to turn Claude Code into a personalized AI system.

engram init

engram bundle

Generates a portable setup package for platforms that don't support CLI (Claude.ai, Claude Desktop, ChatGPT). Downloads as a zip with instructions, personality config, skills, and memory templates.

engram bundle --for "YourName"

engram serve

Starts an MCP (Model Context Protocol) server that exposes your skills and memory as tools any MCP-compatible client can use.

engram serve

engram skill

Manages your skill library -- create new skills from templates, rebuild the skill index, or list what's installed.

engram skill create MyNewSkill
engram skill index
engram skill list

Architecture

Engram organizes AI infrastructure into five composable layers:

Layer 5: Agent Personalities    -- Who your AI is
Layer 4: Skills & Workflows     -- What your AI can do
Layer 3: Hooks & Events         -- When things happen
Layer 2: Memory & History       -- What your AI remembers
Layer 1: Security & Validation  -- What your AI guards against
             |
           MODEL (pluggable -- Claude, GPT, Gemini, local)

Each layer is independent. Use all five or start with just one.

What You Get

After engram init, your ~/.claude/ directory contains:

| File / Directory | Purpose | |------------------|---------| | CLAUDE.md | AI identity, skill registry, global config | | context.md | Your personal context (who you are, what you work on) | | constitution.md | Personality calibration (tunable YAML dials) | | settings.json | Runtime config with hook wiring | | skills/ | 4 starter skills (Research, DoWork, Reflect, HelloWorld) | | hooks/ | 3 lifecycle hooks (security, context loading, session summary) | | MEMORY/ | Structured memory directories |

Everything is plain files. Markdown, YAML, JSON. No database. No cloud service. Human-readable, git-friendly, portable.

Web Bundle Generator

Don't use CLI? Visit the Engram website to generate a setup package through your browser. Fill in your name, tune personality sliders, and download a zip you can upload to Claude Projects or paste into ChatGPT custom instructions.

Documentation

| Document | Description | |----------|-------------| | Getting Started | Installation, setup, first session | | Creating Skills | Build and register custom skills | | Writing Hooks | Event-driven lifecycle hooks | | Architecture Guide | The five-layer system in detail | | Examples | Skills, hooks, and configuration examples | | FAQ | Common questions |

Contributing

Contributions welcome.

1. Fork the repository
2. Create a feature branch (git checkout -b feature/my-feature)
3. Read the relevant spec in specs/ before making changes
4. Follow existing conventions (TitleCase for skills, TypeScript for hooks)
5. Submit a pull request with a clear description

For bugs and feature requests, open an issue on GitHub.

License

MIT -- see LICENSE for details.