Mnemex

Mnemex is an MCP server that equips AI assistants like Claude with human-like memory dynamics using temporal decay, spaced repetition, and local file storage. Developed in Python and also referenced as CortexGraph, this server connects language models directly to persistent local stores, including JSONL files, SQLite databases, and Markdown notes designed for tools like Obsidian. It is designed for developers, researchers, and AI practitioners who require persistent conversational memory without relying on proprietary cloud backends. By tracking memory recency, access frequency, and calculated importance multipliers, Mnemex simulates natural human retention patterns inspired by the Ebbinghaus forgetting curve. Infrequently accessed facts gradually decay and are purged, while frequently referenced items strengthen over time and undergo promotion to long-term Markdown storage. The server also incorporates automated entity extraction and recall detection from standard conversational text. This architecture enables continuous context preservation across separate interactions, ensuring user preferences, technical specifications, and project decisions remain accessible without requiring manual prompt repetition or complex external database configuration.

Category: AI Memory & Context

Tags: jsonl, markdown, memory, persistence, spaced-repetition

Visit Mnemex

How to install and configure Mnemex

1. Ensure Python 3.10 or higher is installed on your local environment. 2. Install the package using pip install cortexgraph or pip install mnemex. 3. Configure your MCP client, such as Claude Desktop, to register the server using your client's configuration file. 4. Optionally customize decay rates and thresholds using environment variables such as CORTEXGRAPH_DECAY_LAMBDA and CORTEXGRAPH_FORGET_THRESHOLD. 5. Follow the specific connection steps and configuration examples in the repository at https://github.com/simplemindedbot/mnemex.

What you can do with Mnemex

• Retaining developer coding preferences, favorite tools, and environment configurations across multiple sessions without repetitive prompting. - Automatically extracting named entities and domain keywords from user messages to reinforce context during technical discussions. - Promoting frequently referenced project decisions and architectural conventions into permanent Markdown files formatted for Obsidian knowledge vaults. - Fading obsolete task details or ephemeral notes naturally through temporal decay algorithms without requiring manual memory cleanup. - Detecting user recall intent automatically during conversations to search and retrieve relevant historical context without explicit tool triggers.

Key facts

• Open Source
• https://github.com/simplemindedbot/mnemex
• AI Memory & Context, Notes & Knowledge Management
• jsonl, markdown, memory, persistence, spaced-repetition

Part of MCP Servers

What can Mnemex do?

Mnemex gives AI assistants a dynamic memory system based on temporal decay and natural spaced repetition. It automatically extracts entities, tracks memory usage, and allows unused details to fade away while reinforcing frequently used facts. Important memories are promoted to permanent Markdown files or SQLite records, preventing context loss across sessions without requiring external cloud databases.

How do I install Mnemex?

Mnemex requires Python 3.10 or higher. You can install it using pip by running pip install cortexgraph or pip install mnemex in your terminal. After installation, add the server to your MCP client configuration file and consult the official repository instructions at https://github.com/simplemindedbot/mnemex for runtime parameters and environment variable settings.

Which MCP clients work with Mnemex?

Mnemex works with any MCP-compliant application that supports standard input and output tool execution. This includes Claude Desktop, Cursor, and custom agent runtimes built with the Model Context Protocol. Once registered, the client interacts with Mnemex tools automatically during standard chat interactions to store, recall, and reinforce information.

Is Mnemex open source?

Yes, Mnemex is open-source software distributed under the GNU Affero General Public License v3.0 (AGPL-3.0). The project serves as a research artifact and reference proof of concept for cognitive architectures and temporal memory systems. The complete source code is publicly accessible on GitHub.

Where does Mnemex store memory data?

All memory data is saved locally on your machine without external cloud dependencies. Short-term memory items are stored in human-readable JSONL files or an SQLite database located in your user configuration directory. Promoted long-term memories are exported as Markdown notes with YAML frontmatter, suitable for use in tools like Obsidian.