Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.memorycrystal.ai/llms.txt

Use this file to discover all available pages before exploring further.

Start with the universal installer. It detects supported clients, asks before configuring each one, and lets you choose cloud or local backend mode.

​
Agent Instructions

Copy and paste this directly into your agent, and it will know what to do: 
Install Memory Crystal for the agent I am using. Run the official Memory Crystal installer for this client and operating system, follow the auth/backend prompts, do not manually edit config files unless the installer fails, then verify that Memory Crystal tools are available. Refer to this page for the exact command to run for this platform and agent: https://docs.memorycrystal.ai/installation
Common target commands: 
curl -fsSL https://memorycrystal.ai/crystal | bash -s -- --targets openclaw
curl -fsSL https://memorycrystal.ai/crystal | bash -s -- --targets hermes
curl -fsSL https://memorycrystal.ai/crystal | bash -s -- --targets claude-code
curl -fsSL https://memorycrystal.ai/crystal | bash -s -- --targets codex-cli
curl -fsSL https://memorycrystal.ai/crystal | bash -s -- --targets factory-droid

iwr https://memorycrystal.ai/install.ps1 -OutFile install.ps1
.\install.ps1 -Targets claude-code
.\install.ps1 -Targets claude-desktop
.\install.ps1 -Targets codex-cli
.\install.ps1 -Targets factory-droid
Hermes Agent is not supported by the native PowerShell installer yet. OpenClaw PowerShell writes generic MCP guidance; use Bash, WSL, or Git Bash for the native OpenClaw plugin install. Codex Desktop requires generic MCP/manual setup until it exposes a stable config location.

​
Universal install

For macOS, Linux, WSL, or Git Bash: 
curl -fsSL https://memorycrystal.ai/crystal | bash
For native Windows PowerShell, use the PowerShell companion when available for your release: 
iwr https://memorycrystal.ai/install.ps1 -UseBasicParsing | iex
The installer can configure:
  • OpenClaw
  • Hermes Agent
  • Claude Code
  • Claude Desktop
  • Codex CLI
  • Factory Droid
  • generic MCP clients
It prompts before changing each detected client and creates backups before config edits.

​
Choose a deployment mode

ModeBest forDetails
Cloud (recommended)Easiest setup, shared memory across devicesUses Memory Crystal’s hosted backend
Local-firstSingle-machine data controlUses Docker Convex on 127.0.0.1 for data while hosted auth/license remains required
External self-hostedAdvanced operatorsUses your own Convex deployment when supported by the installer
See Local-First Setup for Docker requirements, local dashboard URLs, backup behavior, and rollback notes.

​
Single-client installer paths

Use these when you intentionally want one client path instead of the universal prompt flow. 
curl -fsSL https://memorycrystal.ai/install-claude-mcp.sh | bash
See Claude Code.
curl -fsSL https://memorycrystal.ai/install-codex-mcp.sh | bash
See Codex CLI.
curl -fsSL https://memorycrystal.ai/crystal | bash -s -- --targets hermes
See Hermes Agent.
curl -fsSL https://memorycrystal.ai/install-droid-mcp.sh | bash
See Factory Droid.
Use the config shown on the Claude Desktop page when your installer release does not yet mutate Claude Desktop config automatically.

​
Pick your platform

Cloud and external self-hosted modes need a Memory Crystal API key. Local-first mode still needs hosted account/device auth so Memory Crystal can validate license entitlement before provisioning local credentials.
  1. Visit memorycrystal.ai/dashboard
  2. Create or open your account
  3. Go to API Keys
  4. Create a key and save it somewhere safe
The dashboard only shows the key once. Local mode stores local client credentials separately and should never log hosted bearer/device tokens.

​
Verify your setup

For OpenClaw installs: 
openclaw plugins info crystal-memory
openclaw crystal_status
For source checkouts: 
npm run crystal:doctor
For local Docker Convex, also verify the local backend health from the installed artifact or from a repo checkout: 
npm run convex:local:doctor
If local-first setup fails, run the local backend doctor from the installed artifact and verify Docker is running before reinstalling.