Skip to main content
It depends on how you connect:
  • Via the MCP server — works with any MCP-compatible host, including Claude Desktop, Cursor, Windsurf, Claude Code, and Codex.
  • Via the OpenClaw plugin — works specifically with OpenClaw.
The MCP server is the broadest integration path. If your AI host supports MCP, you can connect Memory Crystal to it.
It depends on your setup:
  • Cloud plan — memories are stored on memorycrystal.ai infrastructure.
  • Self-hosted — memories are stored in your own Convex deployment. Your data never leaves your infrastructure. See Self-Hosting.
  • Local mode — memories are stored in a local SQLite database on your machine. No data is sent to any remote server.
The OpenClaw plugin hooks into two lifecycle events:
  • message_received — captures incoming user messages
  • llm_output — captures assistant responses and triggers memory extraction
Every conversation turn is captured automatically. An LLM extraction step then distills durable facts, decisions, and lessons from the raw turn and stores them in long-term memory. You don’t need to manually save anything.
Yes. The MCP server works independently of OpenClaw and connects to any MCP-compatible host.Supported hosts include Claude Desktop, Claude Code, Cursor, Windsurf, and Codex. See the MCP Server integration guide for setup instructions.Note that the MCP server requires you to manually call memory tools (like crystal_remember and crystal_recall), whereas the OpenClaw plugin captures and injects memory automatically.
Channels are namespace prefixes that isolate memory by tenant, client, or agent lane. When you set a channelScope, all memory operations for that plugin instance are stored and retrieved under the {channelScope}:{peerId} namespace.This is useful when running multiple agents or when building multi-tenant applications where different users or clients should have separate memory spaces. See Plugin Options for configuration details.
Yes, in two ways:
  • crystal_forget tool — archive or permanently delete a specific memory from within your AI session
  • Web dashboard — manage, browse, and delete memories at memorycrystal.ai/dashboard
Archiving is a soft delete — the memory is removed from recall but can be reviewed in the dashboard. Permanent deletion removes it entirely.
All of them. Memory Crystal operates at the infrastructure layer — it captures conversation turns and injects context into the system prompt before the model sees the message. The model itself doesn’t need to support anything special.Memory extraction uses an LLM internally, but this is handled by Memory Crystal’s backend and doesn’t affect which model your AI host is using.
Yes. All endpoints enforce per-API-key rate limits. If you hit a rate limit, you’ll receive an error response. Rate limits apply to memory capture, recall, search, and knowledge base operations.If you need higher limits, upgrade your plan or contact support for Enterprise options.