Mnemosyne Is Now a Universal Memory Layer for Any AI Agent

Six months ago, Mnemosyne was "memory for Hermes Agent." A plugin. A dependency. A tool for one framework.

Today it's something bigger.

Mnemosyne is now a universal memory layer that any AI agent can plug into. Hermes, Claude Code, Cursor, Codex CLI, OpenWebUI, OpenClaw, Windsurf — and any platform that speaks MCP — can all use the same memory backend. One SQLite file. One pip install. Zero cloud dependencies.

Memory shouldn't be locked inside a single tool. Your AI agent's history, preferences, and facts should follow you everywhere — not stay behind when you switch editors.

Why This Matters

The AI tools ecosystem is fragmenting fast. Every month there's a new agent framework, a new IDE plugin, a new coding CLI. They all need memory.

But most memory solutions are tied to a specific platform:

mem0 is an API (cloud-dependent)
Honcho is a managed service
SuperMemory is a SaaS product
Letta requires Docker + PostgreSQL

None of them follow you across tools. Switch from Cursor to Windsurf? Your memory stays behind.

Mnemosyne changes that.

What We Built

MCP Server — Works Everywhere

The same MCP server runs on any platform that supports the Model Context Protocol. Config is always the same one-liner:

{`{
  "mcpServers": {
    "mnemosyne": {
      "command": "mnemosyne",
      "args": ["mcp"],
      "env": {}
    }
  }
}`}

Drop this into:

Platform	Config file
Cursor	`.cursor/mcp.json`
Claude Code	`claude.json`
Codex CLI	`.codex/mcp.json`
Windsurf	`.windsurf/mcp_config.json`
Any MCP client	One config line

OpenWebUI Native Tool

OpenWebUI users get a class-based tool with a full settings panel — configure db_path, bank, top_k, and weights directly in the OpenWebUI UI. Drop in a one-line bridge file and it appears in your workspace.

OpenClaw Memory Provider

For OpenClaw agents, Mnemosyne implements the MemoryProvider ABC directly — install mnemosyne-memory[openclaw], add a config line, and the provider is auto-discovered.

Integration Template

Need to add Mnemosyne to a platform we don't support yet? The integration template shows the ~100-line pattern. It's always the same: connect, expose, configure.

What Makes Mnemosyne Different

The comparison speaks for itself:

	Mnemosyne	mem0	Honcho	SuperMemory	Letta
Local-first	✅ SQLite	⚠️ Hybrid	❌ PG+worker	❌ SaaS	❌ Docker+PG
Zero deps	✅ pip only	❌ Qdrant/PG	❌ PG+3 LLMs	❌ SaaS infra	❌ PG+vector
MCP server	✅ Built-in	❌	❌	✅	❌
Multi-platform	✅ 8+ targets	⚠️ 3 adapters	⚠️ 4 adapters	✅ MCP	❌ Agent-only
Open source	✅ MIT	✅ Apache 2.0	⚠️ AGPL	❌ Proprietary	✅ OSS
Integration template	✅ Published	❌	❌	❌	❌

Every other solution assumes you'll stay on their platform. Mnemosyne assumes you won't.

What's Next

Docker image for MCP server (coming today)
Deep OpenWebUI integration — auto-save every chat, memory browser
VS Code extension for browsing memories
Obsidian plugin — memory as markdown notes

Try It

pipx install mnemosyne-memory

Then add the MCP config to your favorite agent. Takes 30 seconds.

Get started on GitHub →