The Problem
Every AI coding agent I've used has the same fatal flaw: it forgets when you close the session.
Open from folder A → works great
Close → Open folder B → "Who are you? What are we working on?"
Forgetting context isn't just annoying. It's a productivity killer.
The Solution: Open World Architecture
I built a memory system for OpenCode that transforms it from a stateless assistant into a persistent collaborator.
What I Created
~/.config/opencode/
├── .global/ ← YOUR WORKSPACE
│ ├── skills/ ← Custom skills
│ ├── memory/ ← Persistent memory
│ └── projects/ ← Work tracking
├── skills/ ← Base skills
└── plugins/ ← System plugins
The magic: .global/ folder survives folder changes.
Key Features
1. Global Memory
Memory persists across ALL folders. Your preferences, decisions, and context follow you everywhere.
2. Sandbox Upgrades
Want to test a new skill? Create it in .global/skills/. It runs without touching the base installation. Broken? Delete it. Base unaffected.
3. Self-Healing
Crash recovery built-in. The agent auto-restores from backup if something breaks.
4. One-Command Install
git clone https://github.com/Kshitiz-Maurya/opencode-enhanced.git
.\install.ps1
# Done!
How It's Different
| Feature | Traditional AI Agent | My Creation |
|---|---|---|
| Memory | Session only | Global (all folders) |
| Upgrades | Risk breaking | Sandbox safe |
| Setup | Manual | One-command |
| Recovery | Manual | Auto-healing |
| Sharing | Copy files | Git clone |
Try It Yourself
# Clone the starter kit
git clone https://github.com/Kshitiz-Maurya/opencode-enhanced.git
# Run install
.\install.ps1
# Restart OpenCode
# Done! Memory persists from any folder
What's Next?
This is v1. I'm testing it myself before public launch. But the concept is solid: an AI that remembers, learns, and grows with you — not just in one project, but across your entire development workflow.
What do you think? Would a persistent-memory AI coding agent help your workflow?
Built with OpenCode. The architecture is MIT licensed — fork it and make it yours.