Mnemosyne Is Now a Universal Memory Layer for Any AI Agent

import Callout from "@/components/Callout"; # Mnemosyne Is Now a Universal Memory Layer 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 (Model Context Protocol) — can all use the same memory backend. One SQLite file. One pip install. Zero cloud dependencies. ## 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 have one thing in common: **they 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. If you 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: - **Cursor** — Add to `.cursor/mcp.json` - **Claude Code** — Add to `claude.json` - **Codex CLI** — Add to `.codex/mcp.json` - **Windsurf** — Add to `.windsurf/mcp_config.json` - **Any MCP client** — One config line Config is always the same pattern: ```json { "mcpServers": { "mnemosyne": { "command": "mnemosyne", "args": ["mcp"], "env": {} } } } ``` ### 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 1-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](https://github.com/AxDSan/mnemosyne/blob/main/docs/integrations/integration-template.md) shows the ~100-line pattern. It's always the same: connect, expose, configure. ## What makes Mnemosyne different | | 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 | ❌ | ❌ | ❌ | ❌ | ## 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 ```bash pip install mnemosyne-memory ``` Then add the MCP config to your favorite agent. It takes 30 seconds. [Get started →](https://github.com/AxDSan/mnemosyne)