███╗   ███╗███╗   ██╗███████╗███╗   ███╗ ██████╗ ███████╗ ██████╗ ██████╗  ██████╗ ███████╗
████╗ ████║████╗  ██║██╔════╝████╗ ████║██╔═══██╗██╔════╝██╔═══██╗██╔══██╗██╔════╝ ██╔════╝
██╔████╔██║██╔██╗ ██║█████╗  ██╔████╔██║██║   ██║█████╗  ██║   ██║██████╔╝██║  ███╗█████╗  
██║╚██╔╝██║██║╚██╗██║██╔══╝  ██║╚██╔╝██║██║   ██║██╔══╝  ██║   ██║██╔══██╗██║   ██║██╔══╝  
██║ ╚═╝ ██║██║ ╚████║███████╗██║ ╚═╝ ██║╚██████╔╝██║     ╚██████╔╝██║  ██║╚██████╔╝███████╗
╚═╝     ╚═╝╚═╝  ╚═══╝╚══════╝╚═╝     ╚═╝ ╚═════╝ ╚═╝      ╚═════╝ ╚═╝  ╚═╝ ╚═════╝ ╚══════╝

MnemoForge CLI — The Inception Engine of the Mnemosyne Neural OS

⚠️ Active beta. APIs may change between minor versions. We run this in production on the Mnemosyne OS monorepo — feedback welcome.

What is MnemoForge?

MnemoForge is the official CLI for the Mnemosyne Neural OS — a sovereign AI operating system built for the next generation of developer-agent collaboration.

It gives AI agents something they fundamentally lack: persistent, versionable, IDE-agnostic memory.

Every time an agent starts a session, it starts from zero. MnemoForge fixes this:

| System | Purpose | |---|---| | Chronicles | Structured memory files — key decisions, sessions, architectural moments | | Resonance Profiles | IDE + Provider + Workspace identity — any agent, any tool | | Local AI Filter | Ollama integration — compress context before MCP injection (token-safe) | | MCP Server | mnemoforge serve — exposes vault tools to any MCP-compatible agent | | Soul Profiles | mnemoforge soul dex — inject behavioral archetypes (Architect, Shipper...) into your IDE | | Canvas Rules | Living ruleset of agent + human rules, persisted in vault, never re-explained |

"Don't just scaffold code. Scaffold intelligence."

Install

npm install -g @mnemosyne_os/forge

Then run:

mnemoforge

Quick Start

# 1. Configure your vault + Resonance profile
mnemoforge chronicle init

# 2. Configure Local AI (Ollama)
mnemoforge config ollama
# → Detects Ollama, lists models, saves your choice

# 3. Open interactive REPL
mnemoforge forge

# 4. Write a chronicle (interactive)
mnemoforge chronicle commit

# 5. Browse your vault
mnemoforge chronicle open

Commands

🧠 Chronicle (Memory Vault)

mnemoforge chronicle init       # configure vault + Resonance profile
mnemoforge chronicle switch     # change active profile (IDE / Provider)
mnemoforge chronicle commit     # write a new chronicle interactively
mnemoforge chronicle open       # interactive vault browser
mnemoforge chronicle list       # list recent chronicles
mnemoforge chronicle sweep      # generate a daily consolidation chronicle

🏗 Canvas (Project Scaffolding)

mnemoforge canvas               # scaffold a new AI-native project

Templates included: api · react-module · agent-service (+ custom)

💬 Prompt Engine

mnemoforge prompt list          # browse & use prompt templates interactively
mnemoforge prompt create        # create a custom reusable prompt template

⚙️ Config & Local AI

mnemoforge config               # settings dashboard (vault, profile, AI)
mnemoforge config ollama        # detect Ollama, select memory filter model
mnemoforge config edit          # edit vault configuration

🔁 Forge (REPL Mode)

mnemoforge forge                # interactive REPL with all commands

🌐 MCP Server (coming v1.4)

mnemoforge serve                # start MCP server on port 3141
mnemoforge serve --port 4000    # custom port

The MCP server will expose write_chronicle() and get_context() tools, letting Antigravity, Cursor, and Claude Code write chronicles natively — with Ollama pre-filtering context to reduce token usage.

Local AI Integration (Ollama)

MnemoForge supports local AI via Ollama for context compression — filtering chronicle content before MCP injection to reduce token consumption.

# Setup
mnemoforge config ollama

# What it does:
# → Pings Ollama at localhost:11434
# → Lists available models (mistral, deepseek-r1, phi3, llama3...)
# → Saves your choice to vault config
# → Used as pre-filter when reading chronicles via MCP

Recommended models for memory compression:

• phi3:mini — fastest, lightweight
• mistral:7b — good balance
• llama3.2:latest — solid general purpose
• deepseek-r1:latest — best reasoning (slower start)

Resonance Profile Architecture

MnemoVault/
  Mnemosyne-OS/            ← Workspace
    CLI/                   ← Resonance Project
      Antigravity/         ← IDE
        Anthropic/         ← Provider
          CHRONICLE-2026-04-05-session.md
          CHRONICLE-2026-04-05-decision.md
    Dashboard/
      Cursor/
        Claude/
          ...

Each profile is completely independent. Switch between Antigravity/Anthropic and Cursor/Claude without losing any history.

Chronicle styles: session · decision · reflection · sweep · narcissus

Neural Coding

MnemoForge implements Neural Coding — a development methodology where:

• The human holds the intention and understands the system
• The agent understands the context and executes without re-explaining everything
• The memory persists outside any IDE, session, or conversation

The code follows the thought. Not the other way around.

→ Read the Neural Coding definition

Roadmap

• [x] chronicle init/commit/list/open/switch/sweep — agent memory vault
• [x] workspace init/show/add-rule — project safety memory
• [x] project init — Resonance Project hierarchy
• [x] canvas — AI-native project scaffolding
• [x] prompt list/create — reusable prompt templates (interactive)
• [x] forge — interactive REPL mode
• [x] config — settings dashboard + Ollama detection + local AI selection
• [x] serve stub — MCP server preview (v1.3.8)
• [ ] serve — live MCP server with write_chronicle + get_context tools (v1.4)
• [ ] Ollama context compression pipeline for MCP (v1.4)
• [ ] Chronicle certification — cryptographic signature (v1.5)

Tech Stack

| Layer | Technology | |---|---| | Runtime | Node.js 18+ | | Language | TypeScript 5 (strict) | | CLI Framework | Commander.js v11 | | Prompts | Inquirer.js v8 | | Styling | Chalk v4 | | Local AI | Ollama REST API |

Documentation

Full docs on GitBook: mnemosyneos.gitbook.io

• Getting Started
• Chronicle System
• Resonance Profiles
• Local AI & Ollama
• MCP Server (upcoming)
• Neural Coding Principles
• Command Reference

License

MIT © 2026 XPACEGEMS LLC — Tony Trochet

About

XPACEGEMS LLC — Independent AI Software Lab
Miami, FL 33122, USA
Founder & Lead Architect: Tony Trochet

Built as part of Mnemosyne Neural OS — a sovereign AI operating system.
Powered by Antigravity (Google DeepMind) · Claude (Anthropic) · Cursor

"Memory is the architecture of intelligence."

⭐ Star on GitHub · 📖 GitBook Docs · 📦 npm · 🐛 Issues