# Ori Mnemos

> Markdown-native persistent memory infrastructure for AI agents. Open-source MCP server that gives any LLM durable, searchable, graph-connected memory across sessions. No cloud. No API keys. Your files, your machine.

Ori Mnemos is an npm package (`ori-memory`) that provides AI agents with persistent memory through the Model Context Protocol (MCP). It stores notes as plain markdown files with YAML frontmatter, builds a local embedding index using Xenova/all-MiniLM-L6-v2 (no API keys needed), and exposes semantic search, graph queries, and note management through MCP tools.

## Install

- Human: `npm i ori-memory` then `ori init` then `ori serve --mcp`
- Agent: Add `{"ori": {"command": "ori", "args": ["serve", "--mcp"]}}` to your MCP client config

## Key Features

- **MCP server** for AI agent memory (works with Claude Code, Cursor, Windsurf, Cline, or any MCP client)
- **Markdown-native storage** — notes are plain .md files with YAML frontmatter, not locked in a database
- **Local embeddings** — Xenova/all-MiniLM-L6-v2 runs locally, no API keys or cloud dependency
- **Three-signal retrieval** — composite vector search + keyword matching + graph-based spreading activation
- **Knowledge graph** — wiki-links (`[[note title]]`) as edges, PageRank for importance scoring
- **Session persistence** — your agent wakes up knowing who it is, what it was working on, and what it learned
- **Git-versionable** — your AI's memory has full version history, diff it, branch it, merge it
- **No vendor lock-in** — if Ori disappears, your files still work in Obsidian, VS Code, or any text editor

## MCP Tools

- `ori_orient` — Session briefing: daily status, reminders, vault health, active goals
- `ori_add` — Create a note in inbox
- `ori_promote` — Promote an inbox note to notes/ with classification, linking, and area assignment
- `ori_validate` — Validate a note against schema
- `ori_query` — Query the vault (orphans, dangling links, backlinks, cross-project)
- `ori_query_ranked` — Full 3-signal retrieval with intent classification and spreading activation
- `ori_query_similar` — Composite vector search (semantic + metadata)
- `ori_query_important` — Notes ranked by PageRank importance
- `ori_query_fading` — Notes losing vitality, candidates for archival or reconnection
- `ori_prune` — Analyze activation topology and identify archive candidates
- `ori_update` — Update agent identity, goals, methodology, daily status, or reminders
- `ori_status` — Vault overview
- `ori_health` — Full diagnostic
- `ori_index_build` — Build or update the embedding index

## Architecture

Notes are stored as markdown files in a flat directory structure. Each note has YAML frontmatter with metadata (description, type, project tags, status, created date). Wiki-links between notes create a knowledge graph that can be traversed and queried. A local SQLite database stores embeddings for semantic search. The MCP server exposes all operations as tools that any MCP-compatible client can call.

## Links

- [Website](https://orimnemos.com)
- [GitHub Repository](https://github.com/aayoawoyemi/Ori-Mnemos)
- [npm Package](https://www.npmjs.com/package/ori-memory)
- [License](https://opensource.org/licenses/Apache-2.0): Apache-2.0