Artemis — AI Engineering CLI

Built by 420.COMPANY · npm · GitHub Created by 420.COMPANY with collaborative engineering support from OpenAI Codex and Anthropic Claude.

Artemis is a full-featured AI engineering CLI built for people who take their workflow seriously. Thirty-plus AI providers. Ninety pre-bundled MCP plugins. Nine hundred and ninety-nine specialized skills. A dual-model brain architecture. A live messaging bridge. And a Spotify integration so your workspace arrives ready — the moment you do.

Table of Contents

• Quick Start
• Installation
• Core Workflows
• Heimdall — Thread Visibility
• Bifrost — Dual-Model Architecture
• MCP Plugins
• 999 Skills
• Messaging Bridge (Bragi)
• Spotify Integration
• Memory System
• Visual Generation (Freya)
• Agent Profiles
• Odin Skill Evolution
• Supported AI Providers
• All Commands Reference
• Permission Modes
• Configuration
• Development
• License
• 中文说明

Quick Start

npm install -g artemis-code
artemis

The CLI guides you through provider setup on first run. Everything else is optional — add what you need, when you need it.

Installation

Requirements: Node.js ≥ 20

npm install -g artemis-code

MCP server dependencies are installed on first use, not at install time. Your npm install -g completes in seconds.

Core Workflows

Artemis ships six named workflows, each a distinct strategy for turning a task description into working code.

/team — Auto-Router (recommended starting point)

Tell Artemis what you need. It analyzes your intent and dispatches to the most appropriate workflow automatically. No need to know which mode to pick.

/team refactor the payment module to support idempotent transactions

/niko — Explore → Build

A two-phase approach: first explore the problem space, gather evidence, and map the unknowns — then build from a position of certainty. Ideal when the best path forward isn't obvious up front.

/niko add multi-tenant support to the auth layer

Phase 1: Research. Reads codebase, identifies dependencies, considers edge cases.
Phase 2: Execution. Builds from findings, not assumptions.

/design — Design First

Produces a full architecture document — data models, interfaces, component breakdown, migration strategy — before a single line of code is written. You review and approve the design. Then Artemis builds it.

/design redesign the notification system to be event-driven

Best for features where a poor early decision compounds into months of technical debt.

/athena — Deep Research + Coordinated Multi-Agent Execution

Deploys multiple independent sub-agents in parallel to research the codebase, then synthesizes findings into a unified execution plan. Built for tasks that span many files, modules, or concerns simultaneously.

/athena audit all authentication flows for OWASP Top 10 vulnerabilities

Roles deployed: planner, researcher (parallel), builder, reviewer

/nidhogg — Adversarial Hardening

Implements a solution, then turns an adversarial critic loose on it. The critic identifies weaknesses, edge cases, and failure modes. Artemis fixes them. Repeat until the solution holds under pressure.

/nidhogg implement the withdrawal transaction handler

The slowest workflow. The most robust output. Use it when correctness is not negotiable.

/contest — Path Debate & Selection

Generates multiple competing approaches to the same problem, debates their trade-offs explicitly, selects the winner, and implements it. Useful when several architecturally distinct solutions are all plausible.

/contest choose between event sourcing and CQRS for our order history feature

Heimdall — Thread Visibility

/heimdall gives you real-time visibility into the agent's execution state: what it's doing, what it's waiting for, and what it needs from you.

/heimdall              # current thread status
/heimdall threads      # list all active threads
/heimdall events       # recent event log
/heimdall follow       # live-stream events in real time
/heimdall upload <file> # pass files into an active thread
/heimdall approve      # unblock a pending approval gate
/heimdall reply <text> # respond to an agent clarification request

When a long-running task pauses for approval or input, /heimdall is how you interact with it without restarting.

Bifrost — Dual-Model Architecture

Most tasks don't need the same model for planning and execution. Bifrost lets you configure two providers independently: one for reasoning, one for speed.

/bifrost   # open dual-model configuration
/mind      # swap active Forge ↔ Raven assignment

Brain model (Raven): High-capability reasoning model — used for planning, analysis, adversarial critique. Think Claude Opus, GPT-4o, DeepSeek-R1.

Execution model (Forge): Fast, code-focused model — handles tool calls, edits, and generation. Think Claude Haiku, GPT-4o Mini, DeepSeek-Coder.

The result: reasoning quality of a large model, with the latency and cost profile of a small one.

MCP Plugins

Ninety MCP servers ship with Artemis, pre-configured and ready to enable. Zero manual setup for HTTP-based servers — they connect on first use.

/mcp list              # browse all 90 servers
/mcp enable <id>       # enable a server
/mcp probe <id>        # test connectivity
/mcp add <command>     # add a custom server
/mcp doctor            # diagnose MCP configuration

Highlights:

| Server | Service | What it unlocks | |--------|---------|-----------------| | cco-github-github | GitHub | Repos, PRs, Issues, Actions | | cco-vercel-vercel | Vercel | Deploy, logs, domains | | cco-firebase-firebase | Firebase | Projects, Firestore, Hosting, Auth | | cco-prisma-prisma-local | Prisma | Schema management, migrations | | cco-notion-notion | Notion | Databases, documents | | cco-slack-slack | Slack | Channels, messages, users | | cco-figma-figma | Figma | Design file access | | cco-atlassian-atlassian | Jira / Confluence | Projects, tickets, docs | | cco-postman-postman | Postman | API collections, test runs | | cco-shopify-shopify-mcp | Shopify | eCommerce operations | | cco-aikido-aikido-mcp | Aikido Security | Vulnerability scanning | | cco-aws-serverless-aws-serverless-mcp | AWS SAM | Serverless deployment | | cco-azure-azure | Azure | Resource management | | cco-sourcegraph-sourcegraph | Sourcegraph | Deep codebase search |

Transport types: 53 streamable HTTP (zero setup), 23 npm/npx, 9 Python/uvx, 4 binary.

999 Skills

Skills are structured knowledge units — pre-built patterns for recurring engineering tasks. Artemis applies them automatically when a task matches.

/skills          # browse the full skill library
/odin list       # inspect learned skills
/odin search <keywords>  # find skills by topic

The Odin layer evolves your skill library over time: after complex multi-step operations, Artemis can capture a new skill from what it just learned — so the next time the same pattern appears, it's faster and more precise.

Messaging Bridge (Bragi)

Send tasks from Telegram, Discord, or WeChat. Artemis executes them on your machine with the same tool access as the terminal — and streams results back to your phone.

/bragi                       # control plane overview
/bragi telegram setup        # configure Telegram bot
/bragi discord setup         # configure Discord bot
/bragi wechat setup          # configure WeChat gateway

The bridge inherits your active permission mode. Set /accept-all and your phone becomes a remote execution terminal.

Spotify Integration

Artemis connects to Spotify so your workspace comes alive the moment you sit down. Search and play music, control playback, switch devices — all without leaving the terminal or your messaging app.

Available tools:

spotify_play_liked       Play your Liked Songs (with optional shuffle)
spotify_search_and_play  Find and play any track or playlist
spotify_play_playlist    Play a named playlist from your library
spotify_resume / pause   Playback control
spotify_skip_next/prev   Track navigation
spotify_set_volume       Volume control (0–100)
spotify_now_playing      Show currently playing track
spotify_set_device       Transfer playback to any device

Via messaging bridge: You can control Spotify directly from Telegram, Discord, or WeChat — ask Artemis to play something and it happens on your machine, wherever you are.

If no active device is found, Artemis automatically wakes the local Spotify app and retries.

Memory System

WordUP — Session Snapshots

Artemis saves compressed memory of your conversation context at key moments. Close the terminal, reopen later, and pick up exactly where you left off.

/wordup          # save a named snapshot
/wordupnow       # force-save immediately
artemis resume --last       # restore most recent snapshot
artemis resume <sessionId>  # restore specific snapshot

Soul File

~/.artemis/soul.md — Write your preferences here. Artemis reads it at startup and applies it to every interaction: tone, communication style, what to avoid, how to present code.

Visual Generation (Freya)

The Freya pipeline adds image and video generation to the standard agent toolset. When a task requires a logo, banner, diagram, or hero image, Artemis generates it directly.

/config visual   # configure visual provider and defaults
/config vision   # configure vision model for image understanding

Supported providers: OpenAI DALL-E, Google Gemini, Grok (xAI), Stable Diffusion, BytePlus, and custom endpoints.

Agent Profiles

Multi-agent workflows like /athena and /nidhogg compose from nine specialist roles, each optimized for a specific function:

| Role | Responsibility | Default Permission | |------|---------------|-------------------| | Planner | Decompose task into sequenced steps | Read-only | | Researcher | Codebase investigation, evidence gathering | Read-only | | Builder | Precise, minimal code changes | Inherits mode | | Reviewer | Find defects, risks, missing coverage | Read-only | | Brainstormer | Diverge → converge on best approach | Read-only | | Arbiter | Judge competing proposals, decide | Read-only | | Architect | System design, scalability, extensibility | Read-only | | Designer | UI/UX, SVG, frontend assets | Inherits mode | | QA | Test strategy, performance, security | Read-only |

Odin Skill Evolution

Odin watches your sessions and learns from them. After a sufficiently complex multi-step operation, it can capture what happened as a structured skill — complete with principles, tool chains, and known edge cases.

/odin list                # view active skills
/odin search <keywords>   # find by topic
/odin capture <name>      # manually save a skill
/odin decay               # run confidence decay on stale skills
/odin remove <id>         # remove a skill

Skills evolve: draft → active → stale → archived. Confidence degrades with disuse and strengthens with reuse.

Supported AI Providers

30+ providers supported out of the box. Configure via /model:config or the setup wizard.

| Provider | Region | Notes | |----------|--------|-------| | Anthropic | Global | Claude 3.5 / 4 family | | OpenAI | Global | GPT-4o, o1, o3 | | Google Gemini | Global | Gemini 2.0 / 2.5 | | DeepSeek | CN / Global | R1, V3, Coder | | Mistral | EU | Codestral, Mixtral | | Groq | Global | Fast inference | | OpenRouter | Global | Meta-provider | | xAI (Grok) | Global | Grok-3 | | Qwen (Alibaba) | CN | Qwen2.5-Coder | | Kimi (Moonshot) | CN | Long context | | BytePlus | CN | Skylark family | | Zhipu (GLM) | CN | GLM-4 | | Baidu (ERNIE) | CN | ERNIE 4.0 | | Minimax | CN | | | 360 AI | CN | | | Lingyi (01.AI) | CN | | | Spark (iFLYTEK) | CN | | | Hunyuan (Tencent) | CN | | | Doubao (ByteDance) | CN | | | Any OpenAI-compatible endpoint | — | Custom base URL |

All Commands Reference

Core
  /help              Help and command list
  /commands          Full command catalog
  /new               Start new session
  /clear             Clear screen / reset history
  /exit              Exit session
  /language          Switch interface language
  /version           Show version

Workflows
  /team              Auto-route to best workflow
  /niko              Explore → Build
  /design            Design first → Implement
  /athena            Multi-agent research + execution
  /nidhogg           Harness engineering loop: build, verify, critique, judge
  /contest           Path debate and selection

Thread Control
  /heimdall          Thread status and control
  /ps                List background runtimes
  /logs <id>         Show runtime logs
  /wait <id>         Wait for runtime to complete
  /attach <id>       Attach to a runtime's session
  /kill <id>         Interrupt a runtime

Model & Providers
  /model [name]      Switch model mid-session
  /model:config      Reopen provider setup wizard
  /providers         Show saved provider profiles
  /bifrost           Dual-model configuration
  /mind              Swap Forge ↔ Raven assignment

Messaging Bridge
  /bragi             Bridge control plane
  /bragi telegram    Telegram sessions and setup
  /bragi discord     Discord sessions and setup
  /bragi wechat      WeChat sessions and setup

Knowledge & Research
  /docs <query>      Documentation lookup
  /research <query>  Deep research shortcut
  /deep-research     Configure deep research engine

Skills & Plugins
  /skills            Browse 999 bundled skills
  /odin              Skill evolution layer
  /mcp               Manage MCP servers
  /plugins           Inspect and run local plugins

Memory & Sessions
  /wordup            Save session snapshot
  /wordupnow         Force-save snapshot
  /sessions          Manage saved sessions
  /history           Recent message history
  /context           Current context snapshot
  /summary           Current session summary

Configuration
  /config            View or update configuration
  /config visual     Visual model setup
  /config memory     Memory enhancement setup
  /artemis-md        Audit project instructions
  /revise-artemis-md Generate instruction improvements
  /doctor            System diagnostic

Tasks & Planning
  /plan              Show execution plan
  /tasks             Task board (add, start, done, block)
  /workflow          Workflow record

Permissions
  /mode <mode>       Set permission mode
  /whosyourdaddy     Enable full autonomy mode
  /evidence          Evidence graph
  /verify            Generate verification plan

Permission Modes

/mode prompt         # (default) confirm before writes and shell commands
/mode read-only      # inspection only — no writes
/mode accept-edits   # auto-approve file edits, confirm shell/network
/mode accept-all     # full autonomy — no confirmations

Or launch directly into a mode:

artemis --accept-all
artemis --read-only

Configuration

Artemis stores all configuration in ~/.artemis/:

~/.artemis/
  config.json          Provider profiles and settings
  trust.json           Trusted workspaces
  sessions/            WordUP session snapshots
  soul.md              Your personal AI personality file
  skills/              Learned Odin skills

Project-level instructions: place an ARTEMIS.md at the root of any repository. Artemis reads it on startup and applies it to all interactions within that workspace. Legacy Artemis.MD, Artemis.md, artemis.md, and .artemis.md files are still accepted for compatibility.

Development

git clone https://github.com/420company/artemis
cd artemis
npm install
npm run run        # run from source
npm run typecheck  # must pass with zero errors
npm run lint       # ESLint
npm run test:all   # run all smoke tests
npm run build      # compile to dist/
npm run release:check  # validate release readiness

See CONTRIBUTING.md for contribution guidelines and docs/RELEASE.md for the GitHub/npm release flow.

License

MIT — open source, free for everyone. Fork it, extend it, make it yours.

See LICENSE for the full text.

中文说明

由 420.COMPANY 构建

Artemis 是一款面向工程师的全功能 AI 编程 CLI，专为认真对待工作流的人设计。三十余家 AI Provider、九十个预装 MCP 插件、九百九十九个专属技能包、双模型脑架构、实时消息桥接——以及一个 Spotify 集成，让你的工作区在你到达的那一刻就已就绪。

我们为 vibe coder 打造了更好的 coding 氛围。得益于 Artemis 的 Spotify 集成，你可以通过 Telegram、Discord 或微信等第三方通讯软件随时控制你的 Spotify——无论身处何地，当你打开工作区时，好的 vibes 早已等候在那里。

快速开始

npm install -g artemis-code
artemis

首次启动时，CLI 会引导你完成 Provider 配置。其余一切按需启用。

核心工作流

/team — 自动路由 (推荐新手起点)

描述你的任务，Artemis 自动判断并分发到最合适的工作流。

/niko — 探索 → 构建

先广泛探索问题空间、收集证据，再从确定性出发构建实现。适用于最佳路径尚不明确的任务。

/design — 设计优先

先生成完整架构文档（数据模型、接口定义、组件拆分、迁移策略），你审核并确认后，再动手写代码。

/athena — 深度研究 + 多 Agent 协作执行

并行部署多个独立子 Agent 研究代码库，汇总发现后制定统一执行计划。适用于跨模块、跨文件的大型任务。

/nidhogg — 对抗性加固

实现方案 → 对抗性审视弱点 → 迭代修复，直到方案在压力下仍然稳健。最慢，但输出最可靠。适用于支付、安全、数据完整性等关键路径。

/contest — 路径辩论与选择

生成多种竞争性实现方案，明确辩论利弊权衡，选出最优方案并实现。

Heimdall — 线程可观测性

/heimdall 让你实时看到 Agent 的执行状态、等待原因和所需输入。

/heimdall              # 当前线程状态
/heimdall follow       # 实时事件流
/heimdall approve      # 解除待审批阻塞
/heimdall reply <内容>  # 回复 Agent 的澄清请求
/heimdall upload <文件> # 向活跃线程传入文件

Bifrost — 双模型架构

规划与执行不需要同一个模型。Bifrost 让你为推理和执行分别配置最适合的 Provider。

/bifrost   # 打开双模型配置
/mind      # 切换 Forge ↔ Raven 分配

脑模型（Raven）： 高能力推理模型，负责规划、分析、对抗性审查。如 Claude Opus、GPT-4o、DeepSeek-R1。
执行模型（Forge）： 快速、代码专项模型，负责工具调用、文件编辑、代码生成。如 Claude Haiku、GPT-4o Mini、DeepSeek-Coder。

效果：大模型的推理质量 + 小模型的速度与成本。

MCP 插件

九十个 MCP 服务器预装就绪，HTTP 类型的服务器零配置即可启用。

/mcp list              # 浏览全部 90 个服务器
/mcp enable <id>       # 启用某个服务器
/mcp doctor            # 诊断 MCP 配置状态

包含：GitHub、Vercel、Prisma、Notion、Slack、Figma、Jira/Confluence、Postman、Shopify、AWS SAM、Azure、Sourcegraph、Aikido Security 等。

消息桥接（Bragi）

通过 Telegram、Discord 或微信发送任务。Artemis 在你的本地机器上执行，并将结果推送回你的手机。与终端完全相同的工具权限。

/bragi telegram setup   # 配置 Telegram 机器人
/bragi discord setup    # 配置 Discord 机器人
/bragi wechat setup     # 配置微信网关

Spotify 集成

Artemis 连接 Spotify，让你的工作区在你就座时就已进入状态。

可通过终端或 Telegram / Discord / 微信 控制：

• 播放喜爱的歌曲或播放列表
• 暂停、继续、切换曲目
• 调节音量、切换播放设备
• 查看当前正在播放的内容

找不到活跃设备时，Artemis 会自动唤醒本地 Spotify 客户端并重试。

记忆系统

WordUP — 会话快照

/wordup          # 保存命名快照
/wordupnow       # 立即强制保存
artemis resume --last   # 恢复最近的快照

Soul 文件

~/.artemis/soul.md — 写下你的个性偏好。Artemis 启动时读取，并应用于所有交互：语气、沟通风格、代码展示方式。

支持的 AI Provider

支持 30+ 家 Provider：Anthropic、OpenAI、Google Gemini、DeepSeek、Qwen（阿里）、Kimi（月之暗面）、字节豆包、百度文心、智谱 GLM、BytePlus（火山引擎）、科大讯飞星火、腾讯混元、Minimax、360 AI、零一万物、Mistral、Groq、OpenRouter、xAI Grok，以及任意 OpenAI 兼容端点。

开发

git clone https://github.com/420company/artemis
cd artemis
npm install
npm run run        # 从源码运行
npm run typecheck  # 必须零错误
npm run build      # 编译至 dist/
npm run release:check  # 发布前完整检查

详见 CONTRIBUTING.md。GitHub/npm 发布流程见 docs/RELEASE.md。

开源协议

MIT — 开源，免费，欢迎所有人。Fork 它，扩展它，让它成为你自己的工具。

欢迎 PR、Issue 和一切形式的贡献。

https://github.com/420company/artemis