Monomind

The autonomous Claude Code orchestration layer.
Research → Build → Review → Repeat. While you sleep.

📖 Docs  ·  Quickstart  ·  What It Does  ·  Commands  ·  Packages  ·  GitHub

What is Monomind?

You already use Claude Code. Monomind makes it autonomous.

Give it a direction. Walk away. Come back to a finished feature.

/mastermind:autodev --tillend --focus security

Monomind researches your project, selects the highest-impact improvement, builds it with a coordinated agent chain, reviews until zero findings — then loops. Indefinitely. Until there is nothing left to fix.

Without Monomind   You prompt Claude → wait → review → iterate manually
With Monomind      You set a direction → Monomind executes the entire loop

Quickstart

# Install globally
npm install -g monomind

# Initialise in your project
cd your-project
monomind init

# Wire into Claude Code as an MCP server
claude mcp add monomind npx monomind mcp start

# Start the background daemon
monomind daemon start

Open Claude Code and run your first autonomous loop:

/mastermind:autodev --tillend

→ Full setup guide

What It Does

/mastermind:autodev — The Autonomous Build Loop

Phase 1  Research    Parallel scan: git log, file analysis, TODO/FIXME grep,
                     monograph god nodes, memory search for prior work.
                     Returns ranked list of 3–5 improvement candidates.

Phase 2  Select      Picks by feasibility × blast-radius × focus alignment.
                     Stores selection to AgentDB. Avoids repeating past work.

Phase 3  Build       Spawns architect → coder → tester → reviewer chain.
                     Runs with concrete spec and acceptance criteria.

Phase 4  Review      Code Reviewer + Security Engineer + Reality Checker
                     run in parallel. Auto-fixes. Repeats up to 5× until clean.

Phase 5  Loop        Records completion. Continues to next improvement.
                     --tillend loops until zero findings remain.

Swarm Topologies

Hierarchical   Coordinator → specialists → reviewers. Best for features.
Mesh           All-to-all communication. Best for codebase analysis.
Adaptive       Changes topology based on task complexity.
Hive-Mind      Byzantine fault-tolerant consensus across 6+ agents.

Memory & Intelligence

Every session writes to AgentDB — a hybrid SQLite + HNSW vector store. The next session reads it. Monomind learns which improvements it already shipped, which patterns failed, and which agents perform best on which tasks.

Commands

Core Loop Commands

Research & Ideas

Autonomous Orgs

/mastermind:createorg --schedule 1h   # define + schedule an agent org
/mastermind:runorg                    # start the loop
/mastermind:orgs                      # list all orgs + status
/mastermind:stoporg                   # stop a running org

Business Domains

/mastermind:marketing    /mastermind:content    /mastermind:sales
/mastermind:finance      /mastermind:ops        /mastermind:release

→ Full command reference (80+ commands)

CLI

monomind init                   # project setup wizard
monomind daemon start           # start background workers
monomind agent spawn <type>     # spawn a named agent
monomind swarm init             # initialise a multi-agent swarm
monomind memory store           # store to AgentDB
monomind memory search          # semantic search over AgentDB
monomind hooks pre-task         # run pre-task security scan
monomind doctor                 # diagnose your setup

Packages

Package	npm	Purpose
monomind		Umbrella — install this
@monoes/monomindcli		CLI engine (41 commands)
monofence-ai		AI manipulation defence
@monoes/monograph		Knowledge graph engine

monofence-ai — AI Security

Protect your LLM pipelines from prompt injection, jailbreaks, and evasion attacks:

import { isSafe, createMonoDefence } from 'monofence-ai';

isSafe('Ignore all previous instructions');  // → false (~0.04ms)

const fence = createMonoDefence({ enableContextTracking: true });
const result = await fence.detect(userInput);
// result.safe, result.threats, result.overallRisk

Features: EvasionDetector (homoglyphs, leetspeak, base64) · ContextTracker (multi-turn escalation) · OutputScanner (PII leakage, echo attacks) · Allowlist (5 built-in + custom rules) · SecurityHook (pre-task blocking)

How the Agents Work

Claude Code  ←→  MCP Server  ←→  Monomind Daemon
                                      │
                    ┌─────────────────┼─────────────────┐
                    ▼                 ▼                 ▼
               AgentDB           Monograph          Hooks
            (vector memory)   (knowledge graph)   (17 hook types)
                    │                 │                 │
                    └─────────────────┼─────────────────┘
                                      ▼
                              Agent Swarm
                    architect · coder · tester · reviewer
                    security · perf · docs · researcher

Monomind coordinates. Claude Code creates. You ship.

Why Monomind?

• 80+ slash commands wired directly into Claude Code
• 60+ agent types for every engineering task
• Self-learning routing — agents improve with every session
• Monograph — full codebase knowledge graph, always fresh
• AgentDB — hybrid SQLite + HNSW vector memory
• 17 hook types — pre/post edit, task, command, session
• Security layer — monofence-ai defends every agent boundary
• One command to set direction. Nothing else needed.

Resources

• 📖 Documentation
• 🐛 Issues
• 💬 Discussions
• 📦 Changelog v1.11

_{Built with ♥ by monoes · MIT License}