Everything below works after installation. Most workflows are automatic. Optional features are clearly marked.

Instant Context Loading

What happens: Start Claude Code. Recent memories load instantly.
🧠 Claude Memory System
────────────────────────────────
πŸ“š Found 8 relevant memories:

β€’ Authentication: JWT with refresh tokens
β€’ Database: PostgreSQL + Prisma ORM
β€’ Frontend: React 18 with TypeScript
β€’ State: Zustand for global state
β€’ Styling: Tailwind CSS + shadcn/ui
β€’ Testing: Vitest + React Testing Library
β€’ API: RESTful with rate limiting
β€’ Preferences: Tabs (4 spaces), strict mode
────────────────────────────────
How it works: The session-start hook pulls the latest saved memories for the current project from the local index. Ask Claude to run a semantic search via the claude-mem MCP tools when you need something older or more specific.

Instant Memory Compression

What happens: Type /compact and the conversation is saved immediately. During installation you can opt-in to also save when /clear runs. Before:
You: How should we handle auth?
Claude: Let's use JWT tokens with...
[100 messages later]
You: /compact
After:
✨ Compressed to memory:
"Project uses JWT authentication with refresh tokens,
15-minute access token expiry, httpOnly cookies for
security, and Redis for token blacklisting."
Zero effort. /compact works out of the box, and /clear compression is available if you enabled it during install.

Instant Recovery (Smart Trash)

What happens: Delete something by accident? It’s still there.
# Oops, deleted important memories
$ claude-mem restore

πŸ—‘οΈ Smart Trash Recovery
────────────────────────────────
Found 3 deleted items:

❯ auth-implementation (2 hours ago)
  database-schema (3 hours ago)
  api-endpoints (yesterday)

Select file to restore:
βœ… Restored auth-implementation.1726096800
Never lose data. Smart Trash keeps everything recoverable. Restore one item at a time to track changes. What happens: Ask Claude to run mcp__claude-mem__chroma_query_documents and it searches your memories by meaning. Typical natural-language prompts find:
  • Authentication flow details
  • JWT implementation fixes
  • Password hashing notes
  • Session management trade-offs
  • Security policies and checklists
No exact matches needed. Call the tool anytime for semantic search.

Session Continuity

What happens: Every session builds on the last. Session 1:
You: Let's use PostgreSQL with Prisma
Claude: Great choice! I'll set up...
Session 2 (next day):
Claude: I see we're using PostgreSQL with Prisma.
        Let me continue with the schema...
No repetition. Claude remembers your decisions.

Status Check

What happens: One command shows everything.
$ claude-mem status

πŸ” Claude Memory System Status Check
=====================================

πŸ“‚ Installed Hook Scripts:
  βœ… pre-compact.js: Found at ~/.claude-mem/hooks/pre-compact.js
  βœ… session-start.js: Found at ~/.claude-mem/hooks/session-start.js
  βœ… session-end.js: Found at ~/.claude-mem/hooks/session-end.js

βš™οΈ  Settings Configuration:
  πŸ“‹ Global: ~/.claude/settings.json
     PreCompact: βœ…
     SessionStart: βœ…
     SessionEnd: βœ…

πŸ“¦ Compressed Transcripts:
  Compressed files: 15
  Archive files: 3

πŸ”§ Runtime Environment:
  βœ… Node.js: v20.10.0
  βœ… Bun: 1.0.25

🧠 Chroma Storage Status:
  βœ… Storage backend: Chroma MCP
  πŸ“ Data location: ~/.claude-mem/chroma
  πŸ” Features: Vector search, semantic similarity, document storage

πŸ“Š Summary:
  βœ… Claude Memory System is installed (Global)

πŸ’‘ To test: Use /compact in Claude Code
Complete visibility. Know exactly what’s happening.

Hook Integration

What happens: Claude Code hooks capture everything automatically.
  • session-start: Loads the latest project memories from the index
  • pre-compact: Compresses when you type /compact
  • session-end: Optionally saves just before /clear (if you enabled save-on-clear)
No manual triggers. Once installed, the hooks run without any extra commands.

Performance

What happens: Everything is optimized for speed.
  • Minified runtime: Smallest possible footprint
  • Lazy loading: Only loads what’s needed
  • Local storage: No network latency
  • Incremental updates: Only processes changes
Zero lag. Designed to be invisible.

Privacy

What happens: Your data stays yours.
  • Local only: Everything stored on your machine
  • No telemetry: Zero data collection
  • Open source: Audit the code yourself
  • Your control: Delete anything, anytime
Complete privacy. No cloud. No tracking. Just local memory.

Project Detection

What happens: Knows which project you’re in. Open /Users/you/my-app and claude-mem:
  1. Detects the project name
  2. Loads project-specific memories
  3. Ignores unrelated context
  4. Maintains project isolation
Smart context. Each project gets its own memory space.

CLAUDE.md Integration

What happens: Adds memory instructions to your CLAUDE.md automatically.
<!-- CLAUDE-MEM QUICK REFERENCE -->
## 🧠 Memory System Quick Reference

### Search Your Memories (SIMPLE & POWERFUL)
- **Semantic search is king**: `mcp__claude-mem__chroma_query_documents(["search terms"])`
- **πŸ”’ ALWAYS include project name in query**: `["claude-mem feature authentication"]` not just `["feature authentication"]`
- **Include dates for temporal search**: `["project-name 2025-09-09 bug fix"]` finds memories from that date
- **Get specific memory**: `mcp__claude-mem__chroma_get_documents(ids: ["document_id"])`

### Storage
- Collection: "claude_memories"
- Archives: ~/.claude-mem/archives/
<!-- /CLAUDE-MEM QUICK REFERENCE -->
Claude knows to search. No manual reminders needed.

Updates

What happens: Stay current automatically.
$ npm update -g claude-mem
$ claude-mem install --force
Always improving. Updates preserve your memories.

The Magic Is No Magic

Everything above happens with zero configuration and zero commands (except the initial install). This is what β€œit just works” actually means.
Start using claude-mem today.