Mnemosyne — Documentation skill for Claude Code
Mnemosyne is a local-first, agent-agnostic memory kernel.
How to install Mnemosyne
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Furinelle/Mnemosyne and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Mnemosyne does
Mnemosyne is a local-first, agent-agnostic memory kernel. Long-term memories are plain Markdown files on your disk, indexed by SQLite FTS5 hybrid retrieval, shared by every agent you use — Claude Code, Codex CLI, Cursor, Hermes, or anything that can speak MCP, run a shell command, or read a file.
Alternatives in Documentation
- 99 Release — MoAI-ADK production release via Enhanced GitHub Flow (CLAUDE.local.md §18) 991 ★
- Wiki Serve — Start a local HTTP server for the built llmwiki site 375 ★
- Cog — 240+ Cognitive architecture for Claude Code -- persistent memory, self-reflection, and foresight via plain-tex 294 ★
README
Mnemosyne
[](https://github.com/Furinelle/Mnemosyne/actions/workflows/ci.yml)
Greek goddess of memory, mother of the nine Muses.
**Mnemosyne is a local-first, agent-agnostic memory kernel.** Long-term memories are plain Markdown files on your disk, indexed by SQLite FTS5 hybrid retrieval, shared by every agent you use — Claude Code, Codex CLI, Cursor, Hermes, or anything that can speak MCP, run a shell command, or read a file.
[中文文档 → README.zh.md](README.zh.md)
Why Mnemosyne
Most agent memory systems make you choose between heavy infrastructure (Postgres + vector DB + Docker), an LLM key on every write, or a single vendor's ecosystem. Mnemosyne refuses all three:
- Files are the source of truth. Every memory is Markdown + YAML
frontmatter. Read it, grep it,
git diffit, edit it by hand. Your memories stay readable forever, with or without Mnemosyne. - Zero heavy dependencies. Python 3.11+ and
portalocker. Retrieval is SQLite FTS5 (CJK-aware hybrid search, optional vector lane and reranker); it degrades gracefully to a pure-stdlib BM25 when FTS5 is unavailable. - No LLM in the loop by default. Auto-distillation is heuristic-first; an LLM engine is opt-in, never required.
- Agent-neutral core, adapters at the edge. One kernel, one store, many doors: MCP tools, a CLI, lifecycle injection events, and direct file access.
Quickstart
git clone https://github.com/Furinelle/Mnemosyne && cd Mnemosyne
pip install -e .
cd /path/to/your/project
python3 -m mnemosyne init # creates .mnemosyne/ + a generic AGENTS.md
python3 -m mnemosyne write --type pitfall --importance 70 \
--title "Example" --content "The auth token expires after 15 minutes."
python3 -m mnemosyne search "auth token"
Global preferences live in `~/.mnemosyne/`, project knowledge in the repository's `.mnemosyne/`.
Integrate any agent
Three access paths, pick
Related Skills
Caveat
Long-term memory layer for Claude Code and Codex — markdown-in-git knowledge base, SQLite FTS5 retrieval, MCP,
Agentic Recall
Long-term memory for agentic tasks. It tells you when it doesn't know. An MCP server for Claude Desktop and Cl
Observational Memory
Local-first shared memory for Claude Code, Codex, OpenCode, Kimi, Grok, Cowork, and Hermes, with bounded obser
Opencode Claude Memory
OpenCode plugin for Claude Code memory: persistent local Markdown memory shared with Claude Code, zero config,
Blink Query
A typed wiki for LLMs — markdown on disk, resolution in the library.
Memory Export
Export memories to markdown for review and backup
Related Agents
Context Retrieval
Context retrieval specialist for gathering relevant memories, code patterns, and framework documentation befor
Python RAG Backend Engineer
Production-grade Python RAG backend engineer specializing in ChromaDB vector storage, sentence-transformers em
Brain Retrieval Engineer
Retrieval engineer — six retrieval strategies (hybrid, graph-enhanced vector, entity-anchored local, guarded T