Getting Started
Introduction
Long-term memory for AI agents — store, recall, and connect knowledge across conversations
HyperMemory
HyperMemory is a persistent knowledge graph memory system for AI agents. Agents store facts, decisions, preferences, and relationships as nodes and hyperedges in a hypergraph, then recall them through hybrid search (BM25, vector similarity, regex) or graph traversal. Memory persists across conversations.
The hosted server at api.hypermemory.io exposes 12 MCP tools (hm_store, hm_recall, hm_find_related, hm_ingest, hm_list_orphans, …) over Streamable HTTP that any MCP-compatible client can call. Most tools also have REST equivalents under /api/v1/memory—see the API Reference for gaps (files and orphans).
Three ways to connect
MCP with OAuth
Paste the server URL into your MCP client. Authenticate via OAuth with PKCE. Works with Claude Desktop, Claude Code, Cursor, Windsurf, OpenClaw, n8n.
CLI with API key
Install the hm CLI from PyPI. Authenticate with an API key or hm login. Pipe-friendly JSON output.
MCP with API key
Generate an API key at app.hypermemory.io and pass it as a Bearer token. For programmatic and server-side use.