Ori Mnemos

Memory infrastructure for AI agents.

Ori is an open source, markdown-native system that gives agents persistent, shared, human-readable memory without a database.

v0.2.0 - TypeScript CLI + MCP server + Claude Code bridge.

The Problem

Most agent workflows restart from near-zero context each session.
That causes repeated decisions, inconsistent behavior, and lost learning.

Tooling has focused on reasoning and tool-calling, but memory is still treated as an addon.
Ori treats memory as first-class infrastructure.

What Ori Does

• Scaffolds a vault with standard memory spaces.
• Captures new memory through a controlled inbox write path.
• Validates note quality against template-defined schema.
• Builds and queries a wiki-link graph.
• Runs full vault diagnostics (schema + links + vitality fade signals).
• Exposes memory operations as MCP tools for agent runtimes.

Core Principles

• Markdown-native: plain files, not opaque storage.
• Agent-native: designed for autonomous loops, not manual note apps.
• Human-legible: inspectable, editable, versionable.
• No lock-in: works with git + filesystem; no required cloud/database.
• Deterministic core: all operations work without an LLM. Enhancement is opt-in.

Installation

npx ori-memory init my-vault

Local Development

For contributors and source builds:

npm install
npm run build
npm link
ori --version

Quick Start

Initialize a vault:

ori init my-vault
cd my-vault

Inspect and diagnose:

ori status
ori health

Capture and validate:

ori add "Memory is the missing layer in agents"
ori validate inbox/memory-is-the-missing-layer-in-agents.md

Query structure:

ori query orphans
ori query dangling
ori query backlinks index
ori query cross-project

Commands

• ori init [dir]
• ori status
• ori health
• ori query <orphans|dangling|backlinks|cross-project> [note]
• ori add <title> [--type <type>]
• ori promote [note] [--dry-run] [--type] [--description] [--links] [--project]
• ori archive [--dry-run]
• ori validate <notePath>
• ori bridge claude-code [--global]
• ori serve --mcp

Claude Code Bridge

Ori's MCP server works with any MCP-compatible client. The Claude Code bridge adds deeper integration: session hooks, validate-on-write, and capture-on-stop.

Project scope:

ori bridge claude-code

Global scope:

ori bridge claude-code --global

Global mode:

• installs hooks under ~/.claude/hooks/ori/
• merges into ~/.claude/settings.json
• is idempotent on repeat runs
• exits cleanly in non-vault directories

MCP Server

Run stdio MCP server:

ori serve --mcp

Client config example:

{
  "mcpServers": {
    "ori": {
      "command": "ori",
      "args": ["serve", "--mcp"]
    }
  }
}

Compatible with Claude Code, Cursor, Windsurf, and any MCP-compliant runtime.

Tools:

• ori_status
• ori_query
• ori_add
• ori_validate
• ori_health
• ori_promote

Response envelope:

{
  "success": true,
  "data": {},
  "warnings": []
}

Vault Structure

vault/
|-- .ori
|-- ori.config.yaml
|-- inbox/
|-- notes/
|-- templates/
|-- ops/
|   |-- sessions/
|   +-- observations/

v0.2 Scope

Included:

• Vault scaffold + config defaults
• Template-based schema validation
• Graph queries (orphans, dangling, backlinks, cross-project)
• Health diagnostics with vitality fade detection
• Inbox capture path (ori add) with optional auto-promotion
• Promotion pipeline: heuristic classification, link detection, graph-based suggestions, footer injection (ori promote)
• Archive workflow for old/isolated notes (ori archive)
• LLM-enhanced promotion via Anthropic API (opt-in; Anthropic is the only supported provider in v0.2)
• Claude Code bridge (project + global)
• MCP transport and tool surface

Not yet included:

• Additional LLM providers beyond Anthropic
• Multi-runtime bridge generators beyond Claude Code
• Hosted sync/distribution layer

Development

npm run lint
npm test
npm run build
npm run check:v01

Positioning

Ori is best understood as memory infrastructure for agent systems, not a note-taking app.

If models are compute and tools are actions, Ori is continuity.

Roadmap

See ROADMAP.md for planned evolution from v0.1 foundations to broader protocol and SDK surface.

License

Apache-2.0