Persistent Memory
for AI Agents

Install in one command.

Pick your platform. Run the command. Approve in your browser. Done.

macOSLinuxWSL/Git BashWindows MCP

Agent Instructions

Copy and paste this directly into your agent, and it will know what to do.

Install Memory Crystal for OpenClaw.
Run this official installer command in a terminal on this machine:
curl -fsSL https://memorycrystal.ai/crystal | bash -s -- --targets openclaw
Follow the installer prompts for authentication, backend mode, and config changes. Do not manually edit agent config files unless the installer fails and asks for a manual fallback.
Refer to this page for the exact command to run for your platform and agent: https://docs.memorycrystal.ai/installation
After installation, verify with: openclaw plugins info crystal-memory; openclaw crystal_status.
Important caveat: PowerShell writes generic MCP guidance for OpenClaw. Use Bash, WSL, or Git Bash for the native OpenClaw plugin install.

Terminal command

$
curl -fsSL https://memorycrystal.ai/crystal | bash -s -- --targets openclaw

Requires OpenClaw installed. Bash installs the native plugin; PowerShell provides the generic MCP path.

Terminal — bash

[ WHY MEMORY CRYSTAL ]

Stop re-explaining.
Start shipping.

Memory Crystal gives your AI agents durable memory across compactions, restarts, and new sessions. Decisions, workflows, and context stay available exactly when your agent needs them.

Works withOpenClawClaude CodeCodex CLICursorAny MCP Client

Knowledge bases • Scoped isolation • Lossless snapshots + trace • Media capture

[ HOW MEMORY CRYSTAL WORKS ]

Two layers. One memory system that stays with your agents.

Memory Crystal handles both the live context your agent needs right now and the persistent memory it should keep forever. It works passively in the background, so context survives compaction, restarts, and switching between tools.

Layer 1

In-session context

Fast, live memory for the conversation happening right now.

  • Captures every turn as you work
  • Semantically enriches messages, entities, and decisions
  • Uses the recall hook to inject the right context before each response

Pipeline

Turn capture → semantic enrichment → recall hook injection

Layer 2

Persistent memory

Durable knowledge that keeps accumulating across sessions, tools, and machines.

  • Syncs memory to Convex in the cloud
  • Survives compaction, restarts, and brand-new sessions
  • Available across tools and machines through MCP
  • Supports scoped knowledge bases for private reference data

Storage

Cloud-synced memory backed by Convex and exposed everywhere through MCP.

Why it feels different

  • Semantic search + knowledge graph, not just text search
  • Immutable knowledge bases for docs, policies, and imported reference material
  • Cross-platform memory for OpenClaw, Claude Code, Codex, Factory, and any MCP client
  • Automatic capture and recall — text, images, audio, video, and PDFs
  • Scoped memory and KB isolation for private client, tenant, and agent lanes

[ MEMORY IN ACTION ]

See what happens when your agent actually remembers.

Real examples across Terminal, Discord, Telegram, Claude Code, and Codex — your agent recalls context no matter where the conversation happens.

macOS terminal showing an AI agent recovering full project context from Memory Crystal after session compaction

Agent recovers full project context after session compaction — no re-explaining needed.

[ HOW IT WORKS ]

From conversation to persistent knowledge in six steps.

Memory Crystal captures, embeds, enriches, and stores your agent's context — then recalls it automatically when needed.

1. Capture

Your agent has a conversation

2. Embed

Memories are semantically embedded

3. Enrich

Knowledge graph extracts relationships

4. Store

Durable cloud storage

5. Recall

Agent queries memory before responding

6. Respond

Agent answers with full context

1. Capture

Your agent has a conversation

Every message, decision, and piece of context flows through the capture hook. Memory Crystal watches for important information — architecture decisions, preferences, project context, action items.

Step 1 of 6

[ EARLY USERS ]

What builders are saying.

We integrated Memory Crystal into our AI platform and it changed everything. Before, the AI started every session essentially cold. Now it wakes up knowing each user's active goals, what progress they've made, what patterns keep showing up. Each of our 30+ users gets their own isolated memory store — completely private. The difference isn't incremental. An AI that remembers is a completely different product than one that forgets.

Jordan

Founder, AI-powered SaaS

I've been building AI agent workflows for my business for a year and the biggest friction was always context loss. Every Monday I'd spend 20 minutes re-orienting my agent on active projects, decisions we'd made, things to avoid. Memory Crystal eliminated that entirely. Now I open a session and it just knows where we are. It's the piece that was missing.

Derek

Entrepreneur & AI builder

What convinced me was the privacy model. Each channel gets its own isolated memory — nothing bleeds between clients, nothing bleeds between contexts. For the kind of sensitive conversations I facilitate, that separation isn't optional. Memory Crystal was the only tool that had it built in from the start, not bolted on. Grateful this exists.

Nathan

Counselor & platform founder

Self-Hosted or Cloud

Get Started — Free

Run Memory Crystal on your own hardware or let us host it. Pick a slug, get your docker compose command, and go.

GET STARTED FREE

[ READY TO STOP LOSING CONTEXT? ]

Install now. Start free.

Free plan includes 500 memories, 500 short-term messages, 7-day message/raw sensory retention, and 3 channels.

Memory Crystal — Persistent Memory for Claude Code, Codex & AI Agents