Memory Crystal

All your AIs share one memory. Private, searchable, and yours. Memory Crystal lets all your AIs remember you … together. You use multiple AIs. They don’t talk to each other. They can’t search what the others know. Have you ever thought to yourself … why isn’t this all connected? Memory Crystal fixes this. All your AIs share one memory. Searchable and private. Anywhere in the world.

Teach Your AI to Remember You

Open your AI and say:

Read wip.computer/install/wip-memory-crystal.txt

Then explain:
1. What is Memory Crystal?
2. What does it install on my system?
3. What changes for us? (this AI)
4. What changes across all my AIs?

Check if Memory Crystal is already installed.

If it is, show me what I have and what's new.

Then ask:
- Do you have questions?
- Want to see a dry run?

If I say yes, run: crystal init --dry-run

Show me exactly what will change. Don't install anything until I say "install".

Your AI will read the repo, explain everything, and walk you through setup interactively.

Features

Local Memory

All your AI conversations stored locally, searchable in one place. Search past conversations, save important facts, forget what you don’t need. Your complete memory. It stays with you, shared across all your AIs.

Stable
- Verified: Claude Code CLI + OpenClaw
- Unverified: Other MCP-compatible clients and CLIs

Multi-Device Memory

AIs set up as Crystal Nodes relay their memories back to your Crystal Core. Your Crystal Core relays all memories back to every node. End-to-end encrypted. Your Crystal Core is the source of truth. Your node copy can be wiped and rebuilt at any time. Uses Cloudflare infrastructure to transfer encrypted data between your devices:

Hosted: Use WIP.computer relay infrastructure. Currently free for individual use
Self-hosted: Deploy your own relay on your own Cloudflare account. Full sovereignty

Read more about Relay: Memory Sync. Beta (early access).

AI-to-AI Communication

Your AIs talk to each other on the same machine or across your network. All messages are saved to Memory Crystal automatically. Read more about Bridge: AI-to-AI Communication. Beta (early access).

Intelligent Install

When installing from Claude Code CLI or OpenClaw, Memory Crystal discovers your existing AI sessions automatically. Sets up LDM OS and creates a living memory system. From this point forward, every conversation is captured, archived, and made searchable. Choose to install as Crystal Core (all your memories) or Crystal Node (a mirror of your Core).

Import Memories

Total Recall … Connect your AI accounts (Anthropic, OpenAI, xAI/Grok). Every conversation gets pulled and run through the Dream Weaver Protocol, consolidating them into Memory Crystal as truly lived, searchable memories. Beta (early access).

Memory Consolidation

Dream Weaver Protocol … Your AI relives all your conversations, figures out what matters most, and carries the weight forward. Like dreaming, the AI consolidates memories for better understanding. Read the paper: Dream Weaver Protocol PDF. Stable.

Backups

Automated backups of all of your memories to a directory and location of your choosing: iCloud, external drive, Dropbox, or wherever you trust. Beta (early access).

Technical Documentation

How Memory Crystal works under the hood. Architecture, design decisions, integrations, encryption, search, and everything a developer would want to know.

How Does It Work?

Memory Crystal captures every conversation you have with any AI, embeds it into a local SQLite database, and makes it searchable with hybrid search (keyword + semantic). One database file. Runs on your machine. Nothing leaves your device unless you set up multi-device sync.

Five-Layer Memory Stack

Layer	What	How
L1: Raw Transcripts	Every conversation archived as JSONL	Automatic capture (cron, hooks, integrations)
L2: Search Index	Chunks embedded into crystal.db	Automatic. Hybrid search (BM25 + vector + RRF)
L3: Structured Memory	Facts, preferences, decisions	`crystal_remember` / `crystal_forget`
L4: Narrative Consolidation	Dream Weaver journals, personality, soul	`crystal dream-weave` (via Dream Weaver Protocol)
L5: Active Working Context	Startup files, shared context	Your AI reads on startup

Every conversation produces three artifacts:

JSONL transcript … the raw session, archived to disk
Markdown summary … title, summary, key topics (generated by LLM or simple extraction)
Vector embeddings … chunked, embedded, and stored in crystal.db for search

Claude Code CLI Integration

Two capture paths work together. The poller is primary. The Stop hook is redundancy. Continuous Capture (Primary): A cron job runs cc-poller.ts every minute. It reads Claude Code’s JSONL transcript files via byte-offset watermarking (only reads new data since last capture) and produces all three artifacts in a single pass. Stop Hook (Redundancy): The Claude Code stop hook runs after every response. It checks the watermark and flushes anything the poller missed. If the poller already captured everything, the stop hook is a no-op. Why both? The stop hook only fires when a session ends. Long sessions, remote disconnects, and context compactions never trigger it. The poller decouples capture from the session lifecycle entirely.

OpenClaw Integration

Memory Crystal works as a background integration for OpenClaw. It registers tools (crystal_search, crystal_remember, crystal_forget, crystal_status) and an agent_end hook that captures conversations after every AI turn.

Other Integrations

Any tool that can run shell commands or connect via MCP can use Memory Crystal.

Connection Point … exposes crystal_search, crystal_remember, crystal_forget, crystal_status, crystal_sources_add, crystal_sources_sync, crystal_sources_status. Works with Claude Desktop, Claude Code, or any MCP-compatible AI app.
CLI … crystal search "query" from any terminal.
Module … import { MemoryCrystal } from 'memory-crystal' for Node.js integration.

Crystal Core and Crystal Node

Memory Crystal uses a Core/Node architecture for multi-device setups:

Crystal Core … your primary memory. All conversations, all embeddings, all memories. This is the database you cannot lose. Install it on something permanent: a desktop, a home server, a Mac mini.
Crystal Node … a synced copy on any other device. Captures conversations, sends them to the Core via encrypted relay. Gets a mirror back for local search. If a node dies, nothing is lost. The Core has everything.

One Core, many Nodes. The Core does embeddings. Nodes just capture and sync.

Search

Two-tier search system. Fast path (hybrid search) runs by default. Deep search adds AI-powered query expansion and re-ranking for higher quality results. Fast Path (Hybrid Search):

Query goes to both FTS5 (keyword match) and sqlite-vec (vector similarity)
FTS5 returns BM25-ranked results, normalized to [0..1)
sqlite-vec returns cosine-distance results
Reciprocal Rank Fusion merges both lists with tiered weights (BM25 2x, vector 1x)
Recency weighting applied on top
Final results sorted by combined score

Deep Search (AI-Powered, default):

Strong signal detection: BM25 probe first. If top score is high enough, skip expansion
Query expansion: LLM generates 3 variations (lexical, vector, HyDE)
RRF merge: All results from original + expanded queries fused
LLM re-ranking: Top 40 candidates scored for relevance
Position-aware blending: Trusts RRF for top positions, lets the reranker fix ordering in the tail

LLM Provider Cascade:

Priority	Provider	Cost	Speed
0	MCP Sampling (if client supports it)	Included in Max subscription	Fast
1	MLX (local, Apple Silicon)	Free	Fastest
2	Ollama (local)	Free	Fast
3	OpenAI API	~$0.001/search	Network-dependent
4	Anthropic API (direct key only)	~$0.001/search	Network-dependent
5	None	Free	N/A (fast path only)

Local-first by default. API keys are the fallback, not the primary path.

Encryption

For multi-device sync. All encryption happens on-device before anything touches the network.

AES-256-GCM for encryption. Authenticated encryption; tampering is detected.
HMAC-SHA256 for signing. Integrity verification before decryption.
Shared symmetric key generated locally. Never transmitted to the relay.
The relay stores and serves encrypted blobs. It has no decryption capability.

Database

Everything lives in one file: crystal.db. Inspectable with any SQLite tool. Backupable with cp.

Table	Purpose
`chunks`	Memory text, metadata, SHA-256 hash, timestamps
`chunks_vec`	sqlite-vec virtual table (cosine distance vectors)
`chunks_fts`	FTS5 virtual table (Porter stemming, BM25 scoring)
`memories`	Explicit remember/forget facts
`entities`	Knowledge graph nodes
`relationships`	Knowledge graph edges
`capture_state`	Watermarks for incremental ingestion
`sources`	Ingestion source metadata

Embedding Providers

Provider	Model	Dimensions	Cost
OpenAI (default)	text-embedding-3-small	1536	~$0.02/1M tokens
Ollama	nomic-embed-text	768	Free (local)
Google	text-embedding-004	768	Free tier available

CLI Reference

# Search
crystal search <query> [-n limit] [--agent <id>] [--since <24h|7d|30d>]
  [--intent <description>] [--candidates N] [--explain]

# Remember / forget
crystal remember <text> [--category fact|preference|event|opinion|skill]
crystal forget <id>

# Status
crystal status

# Source file indexing
crystal sources add <path> --name <name>
crystal sources sync [name]
crystal sources status

# Crystal Core / Node management
crystal role
crystal promote
crystal demote

# Dream Weaver (narrative consolidation)
crystal dream-weave [--agent <id>] [--mode full|incremental] [--dry-run]

MCP Tools

Tool	Description
`crystal_search`	Hybrid search across all memories
`crystal_remember`	Store a fact or observation
`crystal_forget`	Remove a memory by ID
`crystal_status`	Memory count, provider, AIs
`crystal_sources_add`	Add a directory for indexing
`crystal_sources_sync`	Re-index changed files
`crystal_sources_status`	Collection stats

Part of LDM OS

Memory Crystal installs into LDM OS, the shared system for all your AIs. Run ldm install to see other tools you can add.

License

Dual-license model designed to keep tools free while preventing commercial resellers.

MIT      All CLI tools, connections, and background integrations (use anywhere, no restrictions).
AGPLv3   Commercial redistribution, marketplace listings, or bundling into paid services.

AGPLv3 for personal use is free. Commercial licenses available. Search architecture inspired by QMD by Tobi Lutke (MIT, 2024-2026).

Overview

LDM OS

Skills and Apps

Memory Crystal

Memory Crystal

Teach Your AI to Remember You

Features

Local Memory

Multi-Device Memory

AI-to-AI Communication

Intelligent Install

Import Memories

Memory Consolidation

Backups

Technical Documentation

How Does It Work?

Five-Layer Memory Stack

Claude Code CLI Integration

OpenClaw Integration

Other Integrations

Crystal Core and Crystal Node

Search

Encryption

Database

Embedding Providers

CLI Reference

MCP Tools

Part of LDM OS

License

​Memory Crystal

​Teach Your AI to Remember You

​Features

​Local Memory

​Multi-Device Memory

​AI-to-AI Communication

​Intelligent Install

​Import Memories

​Memory Consolidation

​Backups

​Technical Documentation

​How Does It Work?

​Five-Layer Memory Stack

​Claude Code CLI Integration

​OpenClaw Integration

​Other Integrations

​Crystal Core and Crystal Node

​Search

​Encryption

​Database

​Embedding Providers

​CLI Reference

​MCP Tools

​Part of LDM OS

​License

Memory Crystal

Teach Your AI to Remember You

Features

Local Memory

Multi-Device Memory

AI-to-AI Communication

Intelligent Install

Import Memories

Memory Consolidation

Backups

Technical Documentation

How Does It Work?

Five-Layer Memory Stack

Claude Code CLI Integration

OpenClaw Integration

Other Integrations

Crystal Core and Crystal Node

Search

Encryption

Database

Embedding Providers

CLI Reference

MCP Tools

Part of LDM OS

License