Auto Memory

What it does

SANDCODE’s auto memory system persists important context between conversations. Instead of re-explaining your project, preferences, and patterns every time, SANDCODE remembers:

• User preferences — coding style, preferred tools, communication preferences
• Project context — architecture decisions, conventions, important files
• Patterns — recurring tasks, common workflows

Memory is stored as markdown files in ~/.sandcode/projects/<project>/memory/.

How it works

At the start of each conversation, SANDCODE loads relevant memories into context. During conversation, it saves new memories when it learns something worth remembering.

The memory system uses structured types:

• user — information about you (role, preferences, knowledge)
• feedback — guidance on how to approach work (what to do, what to avoid)
• project — ongoing work context, goals, decisions
• reference — pointers to external resources

Memory files

Each memory is a markdown file with frontmatter:

---
name: user-preferences
description: Coding style and tool preferences
type: user
---

Prefers TypeScript over JavaScript. Uses pnpm for package management...

The index file MEMORY.md tracks all memories and is loaded into every conversation.

Disabling auto memory

Auto memory is enabled by default (set in distro defaults). To disable:

{
  "features": {
    "autoMemory": false
  }
}