Skip to content
BranchesTags
 
 

Repository files navigation

English | 简体中文

newtype OS

Multi-Agent Content Team — a multi-agent orchestration system built for content production

Created by huangyihe.

What is this

At its core, newtype OS is an 8-agent multi-layer orchestration system equipped with specialized skills and a memory system, purpose-built for content production.

Think of it as a content team living in your terminal. The Chief (Editor-in-Chief) talks to you, breaks down requirements; the Deputy dispatches execution; 6 specialists each handle their domain — research, fact-checking, retrieval, extraction, writing, and editing.

You ↔ Chief (Editor-in-Chief)
          ↓
      Deputy (Deputy Editor)
          ↓
      Researcher · Fact-Checker · Archivist · Extractor · Writer · Editor

Comparison

@newtype-os/plugin @newtype-os/cli
Nature OpenCode plugin Standalone terminal app
Install bun add @newtype-os/plugin npm install -g @newtype-os/cli
Requires OpenCode Self-contained, no dependencies
Launch opencode nt
Config dir ~/.config/opencode/ ~/.config/newtype/
Project dir .opencode/ .newtype/

Key Differences

@newtype-os/cli isn't "OpenCode with a plugin bolted on." It integrates OpenCode as its core engine and ships as a standalone product. You install it, run nt, and the full 8-agent team is ready. @newtype-os/plugin, on the other hand, is a standard OpenCode plugin that adds the same agent team to an existing OpenCode installation.

The CLI operates in its own configuration namespace: ~/.config/newtype/ for global config, .newtype/ for project-level files. This is completely separate from OpenCode's ~/.config/opencode/ and .opencode/. Installing the CLI won't touch your existing OpenCode setup, and it won't interfere with any OpenCode plugins you already have installed (such as oh-my-opencode). The two coexist without conflict.

From a feature standpoint, both versions deliver the same experience. The 8-agent team, built-in skills, memory system, CLI commands, and MCP integrations are identical regardless of which version you pick. You're not giving anything up by choosing one over the other.

If you want a self-contained setup that works out of the box, or if you already run other OpenCode plugins and prefer clean separation, go with the CLI. If you're already an OpenCode user and simply want to add the newtype agent team to your existing workflow, the plugin is the lighter-weight option.

Supported platforms: macOS (Apple Silicon / Intel), Linux (x64 / ARM64), Windows (x64).

Installation & Usage

newtype OS is designed for both humans and agents.

For Humans

Option 1: newtype CLI (Recommended)

A ready-to-use terminal AI assistant with the full agent team built in.

npm install -g @newtype-os/cli

# Launch the TUI
nt

Option 2: As an OpenCode Plugin

If you already use OpenCode:

cd ~/.config/opencode
bun add @newtype-os/plugin

Edit ~/.config/opencode/opencode.json:

{
  "plugin": ["newtype-profile"]
}

For Agents

After installing newtype CLI, the system ships with a full set of CLI commands (nt research, nt write, nt pipeline, etc.) designed specifically for agent invocation.

Run nt init and newtype will automatically detect locally installed AI development tools and inject skill files, teaching agents how to use these CLI commands:

npm install -g @newtype-os/cli
nt init

Supported tools: Claude Code, Cursor, Copilot, Windsurf, Cline, Roo Code, Zed, Goose, Amazon Q, and more.

Once injected, these agents know when and how to call nt research, nt pipeline, and other commands to augment their own capabilities.

Configuration

Connect Model Providers

Use the /connect command in the TUI to connect model providers (75+ supported), or via CLI:

nt auth login

Configure Agent Models

Option 1: Use the /agent-models command in the TUI to configure models for each agent through the interface.

Option 2: Edit the config file directly.

  • newtype CLI: ~/.config/newtype/newtype-profile.json
  • OpenCode plugin: ~/.config/opencode/newtype-profile.json
{
  "agents": {
    "chief": { "model": "your-preferred-model" },
    "deputy": { "model": "your-preferred-model" },
    "researcher": { "model": "your-preferred-model" },
    "writer": { "model": "your-preferred-model", "temperature": 0.7 }
  }
}

All 8 agents can be configured with independent models and parameters.

Customize Chief's Personality

Create .newtype/SOUL.md (or .opencode/SOUL.md for the OpenCode plugin) to customize Chief's communication style:

/init-soul  # Generate default template

Disable Features

{
  "disabled_agents": ["fact-checker"],
  "disabled_skills": ["super-analyst"],
  "disabled_hooks": ["memory-system"],
  "disabled_mcps": ["sequential-thinking"]
}

Features

Agent Team

Agent Role Responsibility
Chief Editor-in-Chief Your entry point — thought partner + task coordinator
Deputy Deputy Editor Chief's execution layer, dispatches downstream specialists
Researcher Researcher External intelligence gathering, trend discovery
Fact-Checker Verifier Fact verification, source credibility assessment
Archivist Archivist Internal knowledge base retrieval and correlation
Extractor Extractor PDF/image/web content extraction to Markdown
Writer Writer Transforms materials into structured first drafts
Editor Editor Language polish, logic strengthening, style consistency

You only interact with Chief. Chief automatically dispatches other agents as needed.

Built-in Skills & MCP Servers

Agents auto-load the corresponding skill framework when executing tasks:

Skill Command Description
Super Workflow /super-workflow End-to-end content production discipline with mandatory quality gates: standards-first → ideation → outline → draft → review → diagnosis → pre-publish → delivery
Super Analyst /super-analyst 12 analysis frameworks (SWOT, Porter's Five Forces, First Principles, etc.)
Super Writer /super-writer 6 writing methodologies (W.R.I.T.E, AIDA, Storytelling, etc.)
Super Fact-Checker /super-fact-checker Systematic fact-checking + source credibility assessment
Super Editor /super-editor 4-layer editing: structure → paragraph → sentence → word
Super Interviewer /super-interviewer Deep dialogue techniques for requirement mining and brainstorming
Super Obsidian /super-obsidian Obsidian CLI-first vault operations: search, read, create, and manage notes

Super Workflow is the orchestration layer that governs the entire content lifecycle — it defines when to do what and how to verify quality at each stage. Super Writer and Super Editor provide the professional techniques for execution. Chief auto-loads the relevant skill when a task calls for a structured framework. You can also trigger them manually via slash commands. Super Obsidian is auto-loaded when an Obsidian vault (.obsidian/ directory) is detected.

MCP Default Config
websearch (Exa) Enabled None required
sequential-thinking Enabled None required
tavily Disabled Requires api_key
firecrawl Disabled Requires api_key

CLI Commands

Beyond the TUI, newtype provides a set of specialized commands for CLI and agent invocation (all support --json output):

Command Agent(s) Description
nt research [topic] Researcher + Fact-Checker Deep research + source verification
nt write [topic] Writer Multi-style content generation
nt edit [file] Editor 4-layer content refinement
nt fact-check [topic] Fact-Checker Fact-checking and cross-verification
nt analyze [topic] Chief + Researcher Framework-based structured analysis
nt extract Extractor Extract content from files/URLs/images
nt archive <action> Archivist Knowledge base CRUD and semantic search
nt pipeline [topic] Full orchestration Research → Analyze → Write → Fact-check → Edit

Examples:

# Research a topic and output a report
nt research "AI Agent architecture trends 2026" -o research.md

# Write an article based on research
nt write --input research.md --style newsletter -o draft.md

# Full pipeline in one command
nt pipeline "The future of MCP" --style essay --output-dir ./output/

WeChat Integration

newtype CLI has built-in WeClaw integration, bridging your WeChat with the full agent team via the ACP (Agent Client Protocol). Send a message to your WeChat account, and Chief will handle it — research, write, fact-check, all from WeChat.

Command Description
nt wechat setup Download WeClaw binary + scan QR code to link WeChat
nt wechat start Start the WeChat agent bridge (auto-updates before launch)
nt wechat stop Stop the WeChat agent bridge
nt wechat status Show installed version and running state

Quick start:

nt wechat setup    # One-time: download + QR login
nt wechat start    # Start the bridge daemon

The agent bridge runs in the background. WeChat messages are routed to newtype's ACP server, which dispatches them to Chief using your configured model. No API key needed for WeChat — it uses the official iLink API.

Memory System

newtype has a built-in automatic memory system — no manual management needed:

  • Daily summaries: Auto-generated LLM summaries after session idle, stored in .newtype/memory/YYYY-MM-DD.md
  • Full transcripts: Complete conversation history per session in .newtype/memory/full/
  • Long-term memory: Old logs auto-archived to .newtype/MEMORY.md every 7 days
  • Auto-recall: Relevant memories are automatically retrieved and injected into context during conversations

Use /memory-consolidate to manually trigger archival.

Knowledge Base

newtype provides two knowledge base initialization commands to help agents quickly understand your project:

Command Generated File Description
/init AGENTS.md Project-level instruction file defining agent behavior and context for the current project
/init-deep KNOWLEDGE.md Deep knowledge index that auto-scans project structure and generates a detailed codebase map

Links

License

Based on oh-my-opencode, follows SUL-1.0 License.

About

AI Agent Collaboration System for Content Creation

Resources

Readme
License

Contributing

Contributing
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages