Project Memory

Project memory is any system that helps an AI tool remember your project’s history, decisions, and current state across sessions. AI models start fresh every conversation, so without project memory, you’d re-explain your entire project every time. The solution is structured files (like PLAN.md or decision logs) that get loaded at the start of each session, giving the AI continuity it doesn’t naturally have. The best project memory captures what was decided and why, not transcripts of old conversations.

The Simple Version

AI tools have a memory problem. They can hold a conversation brilliantly for an hour, but when the session ends (or the context window fills up), everything resets. Start a new session and the AI acts like it’s never seen your project before.

Project memory is any system that solves this. It’s the bridge between sessions. Instead of the AI forgetting everything, you give it a way to pick up where it left off: what’s been built, what decisions were made, what conventions to follow.

Think of it like a shared document that a new team member reads on their first day. They haven’t been in any of the previous meetings, but after reading the document, they know the project, the goals, the decisions, and the current status. Project memory is that document for AI.

Why It Matters

Without project memory, you hit the same wall every session: re-explaining your project, re-stating your constraints, watching the AI suggest things you already decided against. This is drift in action, and it’s the number one reason AI-built projects stall.

Different tools approach project memory differently:

• Claude Code uses PLAN.md and CLAUDE.md files that get loaded at the start of every session
• Cursor has memory banks that persist across sessions
• SpecStory records entire coding sessions for reference
• CodeMem creates searchable archives of past decisions

The key insight: effective project memory captures decisions, not conversations. “We chose Astro because we need static HTML output” is more useful than a transcript of the conversation where that decision was made.

How It’s Used on This Site

PLAN.md acts as this site’s project memory. It records every completed phase (with one-line summaries), every active phase (with full detail), every architectural decision (framework, hosting, design tokens), and every item that’s been consciously deferred (and why). When Claude Code starts a new session, it reads PLAN.md and immediately knows the project’s full history and current state. This is why the site has stayed coherent across dozens of work sessions, even though each session starts with a fresh context window.