mnemon: An LLM-Supervised Persistent Memory Plugin for DeepSeek Harness
mnemon-dev/mnemon
Mnemon provides LLM-supervised persistent memory for DeepSeek Harness, with graph-based cross-session recall in a single binary, zero API keys, so AI agents never forget.
mnemon is a persistent memory plugin for DeepSeek Harness that gives AI agents long-term, cross-session memory. It uses an LLM-supervised pattern where the host LLM acts as an external supervisor while the binary handles deterministic computation, requiring no separate API key. Core features include a four-graph knowledge store, intent-aware recall, importance decay, and automatic deduplication. It ships as a single local binary with one setup command, compatible with DeepSeek Harness, Claude Code, OpenClaw, and other agent runtimes.
How to Install
dsh plugin --profile web add github:mnemon-dev/mnemon- Category
- Memory & Context
- Platform
- DSH Plugin
- Author
- mnemon-dev
- Distribution
- Plugin
mnemon Key Features
mnemon Repo Summary
What Does It Do?
Mnemon is a persistent memory DSH plugin for DeepSeek Harness that gives AI agents long-term, cross-session memory. It solves the problem of LLM agents forgetting critical decisions between sessions, losing cross-session knowledge, and having early information pushed out of the context window in long conversations. Mnemon uses an LLM-supervised pattern where the host LLM acts as an external supervisor, while the binary handles deterministic computation (storage, graph indexing, search, decay) and the LLM makes judgment calls (what to remember, how to link, when to forget). It requires no separate API key and adds no extra inference cost.
Core Features
- Four-graph knowledge store: builds a real knowledge graph with 87 insights and 2150 edges across four graph types, enabling cross-session knowledge accumulation.
- Intent-aware recall: three primitives—
remember,link,recall—form an intent-native protocol, outputting structured JSON with signal transparency. - Importance decay and automatic deduplication: automatically handles memory timeliness and redundancy, keeping the knowledge base precise.
- Single local binary: one executable, zero API keys, one setup command.
- Compatible with multiple agent runtimes: works with DeepSeek Harness, Claude Code, OpenClaw, and any agent runtime, driving lifecycle via hooks.
How to Use This Plugin?
After enabling Mnemon in DSH, run mnemon setup to auto-detect and configure Claude Code, or use mnemon agency setup to enable the Agency preview for Pi agents. Once set up, memory works automatically in new sessions without extra steps. For Codex, simply run mn. To adjust behavior, modify the behavioral guide or hooks configuration.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: mnemon-dev/mnemon. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.