TL;DR: Memory User Edits in 2 Minutes
The one thing to remember: Memory edits work for FACTS, not BEHAVIORS.
What It Is
memory_user_edits = Sticky notes that persist across all conversations in a Claude Project.
The Commands
| Command | What It Does | Example |
|---|---|---|
view |
See all edits | “Show my memory edits” |
add |
Create new edit | “Remember that I use PostgreSQL” |
remove |
Delete by number | “Remove memory edit #3” |
replace |
Update existing | “Update memory #5 to say January 2026” |
What Works vs. What Doesn’t
| ✅ FACTS (Work) | ❌ BEHAVIORS (Don’t Work) |
|---|---|
| “Database: PostgreSQL 15” | “Always check docs first” |
| “User lives in Milan, Italy” | “Prioritize architecture” |
| “Deadline: January 2026” | “Remember to ask questions” |
Constraints
- 30 edits maximum
- 200 characters per edit
- Project-scoped only (not regular chats)
5-Minute Test
- Open a Claude Project
- Say: “Add a memory edit: User is testing memory_user_edits”
- Start a NEW conversation
- Ask: “What am I testing?”
- Claude should remember ✓
Best Practice
Write atomic facts:
✅ "Backend: PostgreSQL 15 + Knex.js" (34 chars)
❌ "We use a really good database setup" (vague)
The System
Memory edits work best with all four layers:
Project Instructions (HOW to work)
+
Documentation (detailed knowledge)
+
Memory Edits (critical facts) ← You are here
+
Active Steering (real-time corrections)
Results
From 5Levels case study:
- 62% fewer architectural errors
- 100% first-try accuracy (up from 60%)
- Zero context rebuilding needed