What is Root Engine?

Root Engine is a compact agent runtime designed to be easy to read, easy to extend, and fast to ship. It focuses on the fundamentals: agent loop + tools + skills + memory + channels + scheduling—without burying you in frameworks.

If you want a repo you can actually understand end-to-end, modify confidently, and deploy quickly, this is the point.

---

Key Features

• Ultra-Lightweight Core Small, focused agent runtime with clean boundaries between agent logic, tools, and integrations.

• Provider-Driven LLM Support Plug in popular LLM providers (or any OpenAI-compatible endpoint) via a simple provider registry + config.

• Tool Use + Skills System Built-in tools and a skills loader so agents can execute actions, call external capabilities, and stay modular.

• Persistent Memory Optional long-running memory for continuity across sessions.

• Multi-Channel Gateways Run Root Engine through chat platforms and messaging channels (where supported in this repo).

• Scheduled Tasks (Cron) Run proactive reminders, routines, and agent jobs on a schedule.

• MCP Support Connect external tool servers using Model Context Protocol, automatically discovered on startup.

• Security Controls Workspace restrictions and allow-lists to reduce risk when running agents in real environments.

---

Architecture

At a high level:

• A CLI launches an agent or a gateway
• The agent loop runs LLM ↔ tool execution
• A provider registry resolves LLM routing
• Skills extend capabilities cleanly
• Channels handle inbound/outbound messaging
• Cron/heartbeat enable proactive behavior

---

Installation

Requires Python 3.11+

Recommended (global, no venv needed)

# Step 1 — install pipx once per machine (macOS)
brew install pipx && pipx ensurepath

# Step 2 — open a new terminal, then install Root Engine
pipx install root-engine

# Step 3 — onboard
root-engine onboard

Already installed? Use --force to reinstall or upgrade to update:

pipx install root-engine --force   # reinstall current version
pipx upgrade root-engine           # upgrade to latest

pip (traditional)

pip install root-engine

---

Quick Start

Root Engine reads configuration from: ~/.root-engine/config.json

1) Initialize

root-engine onboard

2) Configure your provider + model

Edit ~/.root-engine/config.json and set at minimum:

Provider API key (example: OpenRouter)

{
  "providers": {
    "openrouter": {
      "apiKey": "sk-or-v1-xxx"
    }
  }
}

Default model

{
  "agents": {
    "defaults": {
      "model": "anthropic/claude-opus-4-5"
    }
  }
}

3) Chat

root-engine agent

Or one-shot:

root-engine agent -m "Hello!"

---

Chat Apps

Root Engine can run as a gateway for supported chat platforms (tokens/credentials required). Enable a channel in ~/.root-engine/config.json, then run:

root-engine gateway

Channel Config Examples

Telegram

{
  "channels": {
    "telegram": {
      "enabled": true,
      "token": "YOUR_BOT_TOKEN",
      "allowFrom": ["YOUR_USER_ID"]
    }
  }
}

Discord

{
  "channels": {
    "discord": {
      "enabled": true,
      "token": "YOUR_BOT_TOKEN",
      "allowFrom": ["YOUR_USER_ID"]
    }
  }
}

Slack (Socket Mode)

{
  "channels": {
    "slack": {
      "enabled": true,
      "botToken": "xoxb-...",
      "appToken": "xapp-...",
      "groupPolicy": "mention"
    }
  }
}

---

Configuration

Config file: ~/.root-engine/config.json

Providers

Root Engine uses a provider registry to route models and normalize configuration.

Common provider entries include:

• openrouter
• anthropic
• openai
• deepseek
• groq
• gemini
• minimax
• dashscope
• moonshot
• zhipu
• vllm (local / OpenAI-compatible)
• custom (any OpenAI-compatible API base)

Exact available providers depend on what's included in this repo version.

Custom Provider (Any OpenAI-compatible API)

{
  "providers": {
    "custom": {
      "apiKey": "your-api-key",
      "apiBase": "https://api.your-provider.com/v1"
    }
  },
  "agents": {
    "defaults": {
      "model": "your-model-name"
    }
  }
}

vLLM (local / OpenAI-compatible)

{
  "providers": {
    "vllm": {
      "apiKey": "dummy",
      "apiBase": "http://localhost:8000/v1"
    }
  },
  "agents": {
    "defaults": {
      "model": "meta-llama/Llama-3.1-8B-Instruct"
    }
  }
}

---

MCP (Model Context Protocol)

Root Engine can connect to MCP tool servers and expose them as native tools.

Example config:

{
  "tools": {
    "mcpServers": {
      "filesystem": {
        "command": "npx",
        "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"]
      }
    }
  }
}

Supported transport modes:

• Stdio: command + args
• HTTP: url (remote endpoint)

MCP tools are discovered and registered on startup.

---

Security

For safer local/prod use, restrict tool access to your workspace:

{
  "tools": {
    "restrictToWorkspace": true
  }
}

And restrict who can interact on channels:

{
  "channels": {
    "telegram": {
      "enabled": true,
      "token": "YOUR_BOT_TOKEN",
      "allowFrom": ["YOUR_USER_ID"]
    }
  }
}

---

CLI Reference

Command	Description
root-engine onboard	Initialize config & workspace
root-engine agent	Interactive agent chat
root-engine agent -m "..."	One-shot message
root-engine agent --no-markdown	Plain-text replies
root-engine agent --logs	Show runtime logs
root-engine gateway	Start multi-channel gateway
root-engine status	Show runtime/config status
root-engine channels status	Show channel status
root-engine cron add ...	Add scheduled job
root-engine cron list	List scheduled jobs
root-engine cron remove <id>	Remove scheduled job

Interactive mode exits: exit, quit, /exit, /quit, :q, or Ctrl+D.

---

Scheduled Tasks (Cron)

# Add a job
root-engine cron add --name "daily" --message "Good morning!" --cron "0 9 * * *"
root-engine cron add --name "hourly" --message "Check status" --every 3600

# List jobs
root-engine cron list

# Remove a job
root-engine cron remove <job_id>

---

Docker

Compose

docker compose run --rm root-engine-cli onboard
vim ~/.root-engine/config.json
docker compose up -d root-engine-gateway
docker compose run --rm root-engine-cli agent -m "Hello!"
docker compose logs -f root-engine-gateway
docker compose down

Docker

docker build -t root-engine .

docker run -v ~/.root-engine:/root/.root-engine --rm root-engine onboard
vim ~/.root-engine/config.json

docker run -v ~/.root-engine:/root/.root-engine -p 18790:18790 root-engine gateway
docker run -v ~/.root-engine:/root/.root-engine --rm root-engine agent -m "Hello!"
docker run -v ~/.root-engine:/root/.root-engine --rm root-engine status

---

Project Structure

root_engine/
├── agent/          # Core agent logic
│   ├── loop.py     # Agent loop (LLM ↔ tool execution)
│   ├── context.py  # Prompt builder
│   ├── memory.py   # Persistent memory
│   ├── skills.py   # Skills loader
│   ├── subagent.py # Background task execution
│   └── tools/      # Built-in tools
├── skills/         # Bundled skills
├── channels/       # Chat channel integrations
├── bus/            # Message routing
├── cron/           # Scheduled tasks
├── heartbeat/      # Proactive wake-up
├── providers/      # LLM providers
├── session/        # Conversation sessions
├── config/         # Configuration schema + loader
└── cli/            # CLI commands