Skip to content
AI Developer Tools Open Source

Groundwork Lite: Giving Your AI Assistant a Memory

David Balzan·March 1, 2026·3 min read

I've lost count of how many times I've started a coding session and had to re-explain the entire project to my AI assistant.

"What are we building again?" "Where did we leave off?" "What's already done and what's next?"

The assistant is incredibly capable — it just has no memory between sessions. So every time you open a new chat, you're back to square one. That friction adds up fast.


The Backstory

A while ago I built a much bigger system for myself — a full development methodology with 11 slash commands, templates, decision logs, and a fairly detailed playbook. It worked great for my own projects.

But when I tried showing it to other people, the reaction was almost always the same: "This is… a lot."

So I took a step back and did what any sensible builder would do: I stripped it down to the five commands I actually use every single day. That's how Groundwork Lite was born — the essential core, nothing more.


What Groundwork Lite Actually Is

It's a clean production-ready monorepo starter (React 19 + Hono + Turborepo) with a tiny AI workflow layer on top.

The whole idea is simple: five plain slash commands that read and write normal Markdown files. That's it. Your AI assistant suddenly remembers the project, the current phase, what's done, and what's next — without you having to repeat yourself.

It works with Claude Code, Cursor, and VS Code Copilot. No plugins, no extensions, no lock-in.


How It Works

The magic is surprisingly simple.

Instead of relying on chat history (which gets messy and expensive), everything important lives in a few small Markdown files that both you and the AI can read and update.

The loop looks like this:

/start-session → you code → /check-task/update-focus → next day → /start-session again

  • CURRENT_FOCUS.md remembers exactly where you left off
  • PRODUCTION_ROADMAP.md keeps the big picture alive
  • Phase folders hold detailed task checklists

It feels like your AI finally has a notebook it actually checks.


The 5 Commands

CommandWhat it doesWhen you use it
/kickstartSets up your project docs and roadmapFirst time on a new project
/plan-phaseCreates a clear task list for the current phaseStart of every new phase
/start-sessionLoads everything the AI needs to know right nowBeginning of every coding session
/check-taskMarks tasks complete and updates progressAfter finishing something
/update-focusSaves what you got done and what's nextEnd of a session

That's literally the entire workflow.


The Stack

It comes with a solid, modern monorepo out of the box:

  • React 19 + Vite 6 (frontend)
  • Hono 4 (lightweight API)
  • Turborepo 2 (monorepo tooling)
  • TypeScript 5 + Tailwind 4

Everything is ready to go — no weird config, no missing pieces.


Why Plain Markdown Files?

Because they're the most reliable memory we have.

  • The AI reads them quickly (low token cost)
  • You can open and edit them yourself anytime
  • They work in every IDE and every tool
  • They live in git, so your progress is versioned and safe

No databases, no special formats, no lock-in. Just files.


Try It

It takes about two minutes to get running:

git clone https://github.com/davidbalzan/groundwork-ai-lite.git
cd groundwork-ai-lite
pnpm install && pnpm dev

Check out the repo on GitHub →