CLI-JAW

Stop juggling AI coding tools. Use all of them at once.

English / 한국어 / 中文 / 日本語

---

What is CLI-JAW?

CLI-JAW is an open-source platform that unifies every AI coding CLI you already use — Claude, Codex, Gemini CLI, OpenCode, Copilot CLI — into one assistant with one memory and one dashboard. Your main CLI calls the others as "employees" (sub-agents), so you stop copy-pasting between heavy apps and start giving orders from a single place.

It routes through the subscriptions you already pay for. No per-token API billing. No rate limits on top of what you already have.

One developer is managing 100k+ lines of code with this thing. It's basically a super-app: multi-instance manager + mini-Obsidian notes + Kanban board + browser automation, running locally on your machine.

---

What's New in 2.0.0

This is the general availability release. Here's what landed:

Feature	What it means
Manager Dashboard	A super-app control plane. See every running JAW instance, start/stop/restart them, preview live Web UIs, inspect settings — all from one browser tab
Notes Workspace	A mini-Obsidian baked into the dashboard. Folders, rename/move, split editor+preview, KaTeX math, Mermaid diagrams, syntax-highlighted code blocks
Kanban Board	Drag running instances into lanes. Card editor for tracking what each instance is working on
Employee System	Your main CLI dispatches other CLIs as workers. "Fix the frontend" goes to OpenCode. "Update the API" goes to Codex. You approve the results
Responsive Mobile Layout	Sidebar overlay, touch-friendly controls. Manage your fleet from your phone

Plus: explicit PABCD orchestration (/orchestrate, /pabcd, /continue), 230+ reference skills, SQLite FTS5 memory search, cron/every heartbeat jobs, and Computer Use desktop automation.

---

Recent Architecture Updates

Area	Update
PABCD	Worklog resume is explicit: use /continue. Natural language “continue” stays a normal prompt
Gemini CLI	Full-access Gemini runs use --skip-trust --approval-mode yolo plus --include-directories for OS home access; WSL also includes the Windows user home when discoverable
Messaging	/api/channel/send is the canonical Telegram/Discord outbound path; legacy channel-specific endpoints remain
Heartbeat	Jobs support every and cron schedules with optional IANA time zones
Browser web-AI	Runtime diagnostics/orphan cleanup, persistent tab lifecycle, session reattach, and ChatGPT/Gemini/Grok vendor paths are documented in structure/
OfficeCLI	Office skills now treat HWP/HWPX separately: HWPX is a stable primary path, while binary HWP is rhwp sidecar-gated through officecli hwp doctor --json and officecli capabilities --json
Release gates	npm run gate:all runs named release/doc parity checks when developing locally

---

Two-Line Install

Existing users who already manage Codex, Gemini, Claude, MCP servers, OfficeCLI, or other tools can run a safe update first. Safe install creates ~/.cli-jaw only, skips optional tool/runtime setup, and lets you run jaw init later when you want interactive setup.

# macOS/Linux safe update
JAW_SAFE=1 npm install -g cli-jaw

# Windows PowerShell safe update
$env:JAW_SAFE="1"; npm install -g cli-jaw

Normal install performs the full setup:

npm install -g cli-jaw
jaw dashboard

That's it. Open http://localhost:3457. Requires Node.js 22+.

jaw serve also works if you prefer the classic server mode.

Auto-start on boot: jaw service install (auto-detects systemd / launchd / Docker).

---

Platform Setup

macOS — one-click script

1. Open Terminal (Cmd + Space → type Terminal)
2. Paste and hit Enter:

curl -fsSL https://raw.githubusercontent.com/lidge-jun/cli-jaw/master/scripts/install.sh | bash

3. Authenticate (see below) and launch:

jaw dashboard

Windows / WSL — one-click script

Step 1: Install WSL (PowerShell as Admin)

wsl --install

Restart, then open Ubuntu from the Start Menu.

Step 2: Install CLI-JAW

curl -fsSL https://raw.githubusercontent.com/lidge-jun/cli-jaw/master/scripts/install-wsl.sh | bash

Step 3: Reload and start

source ~/.bashrc
jaw dashboard

The installer uses the user-local npm prefix (~/.local) and writes ~/.local/bin to both ~/.bashrc and ~/.profile, so new Ubuntu shells can find jaw and the bundled CLI tools.

The WSL script is the integrated Linux setup path: it verifies jaw --version, requests bundled CLI tools in strict mode, installs OfficeCLI, and verifies officecli --version before reporting success. Browser support is checked after Chromium/Windows Chrome detection; if no runnable browser is found, the installer prints a warning instead of pretending web-ai/browser automation is fully ready.

WSL Troubleshooting

Problem	Fix
unzip: command not found	Rerun the installer
jaw: command not found	Run source ~/.bashrc or export PATH="$HOME/.local/bin:$PATH"
officecli: command not found	Rerun the WSL installer or bash "$(npm root -g)/cli-jaw/scripts/install-officecli.sh"
Permission errors	sudo chown -R $USER $(npm config get prefix)

Linux

You know what to do.

npm install -g cli-jaw
jaw dashboard

Generic Linux npm install -g cli-jaw keeps optional helper installs best-effort for compatibility. Use the WSL one-click script when you want the stricter integrated setup guarantees, or verify manually with:

jaw --version
officecli --version
jaw doctor

---

Authenticate

You only need one. Pick whichever subscription you already have:

# Free options
copilot login        # GitHub Copilot (free tier available)
opencode             # OpenCode — free models available

# Paid (monthly subscription you already pay for)
claude auth login    # Anthropic Claude Max
codex login          # OpenAI ChatGPT Pro
gemini               # Google Gemini Advanced

Check everything at once: claude auth status && jaw doctor

Example jaw doctor output

🦈 CLI-JAW Doctor — 12 checks

 ✅ Node.js        v22.15.0
 ✅ Claude CLI      installed
 ✅ Codex CLI       installed
 ⚠️ Gemini CLI      not found (optional)
 ✅ OpenCode CLI    installed
 ✅ Copilot CLI     installed
 ✅ Database        jaw.db OK
 ✅ Skills          32 active, 194 reference
 ✅ MCP             3 servers configured
 ✅ Memory          structured/ exists
 ✅ Server          port 3457 available

---

The Dashboard

The dashboard is the main way to use CLI-JAW 2.0. It's a local web app that acts as command center for all your AI instances.

Instance Management

The navigator panel shows every JAW instance — running, stopped, or offline. For each one you can:

• Start / Stop / Restart with one click
• See the active CLI engine, model, port, and working directory
• Use perm mode for persistent instances that survive reboots
• Launch new instances pointed at different project directories

Live Preview

Select any running instance and its Web UI embeds directly in the dashboard via a preview proxy. Chat with your AI, see streaming responses, drag-and-drop files — all without leaving the manager.

Notes

A markdown vault lives inside the dashboard. Think mini-Obsidian:

• Folder tree with create/rename/move/delete
• Raw, split, and preview editing modes
• Dirty-state markers so you don't lose unsaved work
• KaTeX math rendering, Mermaid diagrams, syntax-highlighted code blocks

Kanban Board

Drag instance cards into lanes (Backlog / Ready / Active / Review / Done). Each card links to the live instance and has an editor for notes. Good for tracking what each AI session is working on when you're running multiple instances.

Settings Inspector

View and edit runtime settings for any selected instance: active CLI, model, reasoning effort, permission mode, working directory, employees, skills, MCP servers.

---

How the Employee System Works

This is the core idea: your main CLI calls other CLIs as workers.

You talk to one AI (the "Boss"). When it needs specialized work, it dispatches tasks to employees — each running their own CLI with their own model:

You: "Fix the frontend styling and update the API endpoint"

Boss (Claude) thinks...
  ├── Dispatches to Frontend employee (OpenCode) → "Fix the CSS grid layout in dashboard.tsx"
  ├── Dispatches to Backend employee (Codex)     → "Update /api/users to return pagination metadata"
  └── Synthesizes both results for you

# Under the hood, it's one command:
jaw dispatch --agent "Frontend" --task "Fix the CSS grid layout in dashboard.tsx"

Employees are other AI CLIs configured in your settings. Each has its own session, its own model, its own context. The Boss reviews their output before presenting it to you.

Employees vs. Sub-agents

These are different things:

	Employees	Sub-agents
What	Other AI CLIs (Codex, OpenCode, etc.) configured as workers	Built-in parallel task tool within a single CLI
When	Multi-specialist work across different codebases or domains	Internal research, file reads, parallel analysis
How	jaw dispatch --agent "Name" --task "..."	Automatic — the CLI spawns them internally

Use employees for "Frontend does CSS, Backend does API." Use sub-agents for "read these 5 files in parallel before deciding."

---

Five AI Engines

No per-token API billing. Route through subscriptions you already pay for.

CLI	Default Model	Auth	Cost
Claude	opus-4-6	claude auth login	Claude Max subscription
Codex	gpt-5.5	codex login	ChatGPT Pro subscription
Gemini	gemini-3.1-pro-preview	gemini	Gemini Advanced subscription
OpenCode	minimax-m2.7	opencode	Free models available
Copilot	gpt-5-mini	copilot login	Free tier available

Fallback chain: if one engine is rate-limited, the next picks up. Configure with /fallback [cli1 cli2...].

OpenCode wildcard: connect any model endpoint — OpenRouter, local LLMs, any OpenAI-compatible API.

Switch engines live: /cli codex. Switch models: /model gpt-5.5. Works from Web, Terminal, Telegram, or Discord.

---

PABCD Orchestration

For complex tasks, CLI-JAW uses a 5-phase workflow. You approve every transition.

P (Plan) → A (Audit) → B (Build) → C (Check) → D (Done) → IDLE
   ⛔          ⛔          ⛔         auto        auto

Phase	What happens
P — Plan	Boss writes a diff-level plan. Stops for your review
A — Audit	Read-only worker verifies the plan is feasible (imports exist, signatures match)
B — Build	Boss implements. Read-only worker verifies the result
C — Check	Type-check (tsc --noEmit), docs update, consistency check
D — Done	Summary of all changes. Returns to idle

State is DB-persisted and survives restarts. Workers cannot modify files — only verify. Activate with jaw orchestrate, /orchestrate, or /pabcd; resume an active worklog explicitly with /continue.

---

Memory

Three layers, each covering a different recall horizon.

Layer	What it stores	How it works
History Block	Recent session context	Last 10 sessions, max 8000 chars, scoped to working directory. Injected at prompt start
Memory Flush	Structured knowledge from conversations	Triggered after threshold (default 10 turns). Extracts episodes, daily logs, semantic notes as markdown
Soul + Task Snapshot	Identity and semantic recall	Core values, tone, boundaries. FTS5 index returns up to 4 semantically relevant hits per prompt

All three layers feed into the system prompt automatically. Memory is searchable:

jaw memory search "how did we set up the API auth?"

---

Skills

230+ skills covering dev workflows, office documents, automation, and media.

Category	Skills	What they cover
Office	pdf, docx, xlsx, pptx, hwp	Read, create, edit documents. HWPX is the stable Korean document path; binary HWP is rhwp-backed and capability-gated via OfficeCLI
Automation	browser, vision-click, screen-capture, desktop-control	Chrome CDP, AI-powered coordinate click, macOS screenshots, Computer Use
Media	video, imagegen, lecture-stt, tts	Remotion video, OpenAI image generation, lecture transcription, text-to-speech
Integration	github, notion, telegram-send, memory	Issues/PRs/CI, Notion pages, Telegram media delivery, persistent memory
Visualization	diagram	SVG diagrams, charts, interactive visualizations rendered in chat
Dev Guides	dev, dev-frontend, dev-backend, dev-data, dev-testing, dev-pabcd	Engineering guidelines injected into agent prompts

Reference skills live in skills_ref/ and install into the active runtime on demand; active skills are loaded from the user runtime home.

jaw skill install <name>    # activate a reference skill
jaw skill list              # see what's available

---

Browser & Desktop Automation

Capability	How it works
Chrome CDP	Navigate, click, type, screenshot, evaluate JS, scroll, press keys — full DevTools Protocol control
Vision-click	Screenshot the screen → AI extracts target coordinates → clicks. jaw browser vision-click "Login button"
Computer Use	Desktop app automation via Codex Computer Use MCP. Use Safari for localhost and it feels like the Codex app
Web-AI vendors	jaw browser web-ai --vendor chatgpt|gemini|grok with session lifecycle, diagnostics, and source-audit/answer-artifact support where implemented
Diagram Skill	Generate SVG diagrams and interactive visualizations, rendered inline in chat

Computer Use lets you control any macOS app — Finder, Safari, System Settings, Xcode — through natural language. Point it at your localhost dev server in Safari and you get a full visual testing loop.

---

Messaging

Telegram

📱 Telegram ←→ 🦈 CLI-JAW ←→ 🤖 AI Engines

Text chat, voice messages (auto-transcribed via multi-provider STT), file/photo upload, slash commands (/cli, /model, /status), scheduled task delivery via every/cron heartbeat jobs.

Setup (3 steps)

1. Message @BotFather → /newbot → copy the token
2. jaw init --telegram-token YOUR_TOKEN or use Web UI settings
3. Send any message to your bot. Chat ID is auto-saved on first message

Discord

Same capabilities as Telegram — text, files, commands. Channel/thread routing, canonical /api/channel/send, and forwarder support for agent result broadcast. Setup via Web UI settings.

Voice & STT

Voice input works on Web (mic button), Telegram (voice messages), and Discord. Providers: OpenAI-compatible, Google Vertex AI, or any custom endpoint.

---

MCP

Model Context Protocol lets AI agents use external tools. CLI-JAW manages MCP config for all five engines from one file.

jaw mcp install @anthropic/context7
# → syncs to Claude, Codex, Gemini, OpenCode, Copilot config files simultaneously

No more editing five different JSON files. Install once, every engine gets it.

jaw mcp sync       # re-sync after manual edits

---

CLI Commands

# Core
jaw dashboard                     # launch manager dashboard
jaw serve                         # start server (http://localhost:3457)
jaw chat                          # terminal TUI
jaw doctor                        # 12-point diagnostics

# Instances
jaw clone ~/project               # clone instance to new directory
jaw --home ~/project serve --port 3458  # run second instance
jaw service install               # auto-start on boot

# AI & Orchestration
jaw dispatch --agent "Backend" --task "..."  # dispatch employee
jaw orchestrate                   # enter/control PABCD workflow
# in chat: /continue               # explicit worklog/PABCD resume

# Skills & MCP
jaw skill install <name>          # activate a skill
jaw skill list                    # list available skills
jaw mcp install <package>         # install MCP → syncs to 5 engines
jaw mcp sync                      # re-sync MCP configs

# Memory
jaw memory search <query>         # search across all memory layers
jaw memory save <file> <content>  # save to structured memory

# Browser
jaw browser start                 # launch Chrome (CDP)
jaw browser fetch "https://example.com" --json --trace  # adaptive URL reader
jaw browser snapshot              # capture page state
jaw browser vision-click "Login"  # AI-powered click

# Maintenance
jaw reset                         # full reset

---

Multi-Instance & Docker

Multi-Instance

Run isolated instances with separate settings, memory, and database:

jaw clone ~/my-project
jaw --home ~/my-project serve --port 3458

Each instance is fully independent — different working directory, different memory, different MCP config. The manager dashboard sees them all.

Docker

docker compose up -d       # → http://localhost:3457

Non-root jaw user, Chromium sandbox enabled. Data persists in jaw-data named volume.

Docker details

# Dev build
docker build -f Dockerfile.dev -t cli-jaw:dev .
docker run -d -p 3457:3457 --env-file .env cli-jaw:dev

# Pin version
docker build --build-arg CLI_JAW_VERSION=2.0.0 -t cli-jaw:2.0.0 .

# If Chromium sandbox fails
docker run -e CHROME_NO_SANDBOX=1 -p 3457:3457 cli-jaw

---

Development

npm run build          # tsc → dist/
npm run dev            # tsx server.ts (hot-reload)
npm test               # native Node.js test runner
npm run gate:all       # named release/docs parity gates

Architecture details: ARCHITECTURE.md · Test coverage: TESTS.md · Internal structure docs: structure/

---

How It Compares

	CLI-JAW 2.0	Hermes Agent	Claude Code
Model access	5 CLIs via OAuth (Claude Max, ChatGPT Pro, Copilot, Gemini) + OpenCode wildcard	API keys (OpenRouter 200+, Nous Portal)	Anthropic only
Cost model	Monthly subscriptions you already pay for	Per-token API billing	Anthropic subscription
Primary UI	Manager dashboard + Web PWA + Mac app + TUI	TUI only	CLI + IDE plugins
Dashboard	Multi-instance manager, Kanban, Notes workspace	None	None
Messaging	Telegram (voice) + Discord	Telegram/Discord/Slack/WhatsApp/Signal	None
Memory	3-layer (History/Flush/Soul) + SQLite FTS5	Self-improving loop + Honcho	File-based auto-memory
Multi-agent	Employee system (dispatch other CLIs) + PABCD	Subagent spawn	Task tool
Browser automation	Chrome CDP + vision-click + Computer Use	Limited	Via MCP
Execution	Local + Docker	Local/Docker/SSH/Daytona/Modal	Local
Skills	230+ bundled	Self-creating + agentskills.io	User-configured
i18n	English, Korean, Chinese, Japanese	English	English

---

Troubleshooting

Problem	Solution
cli-jaw: command not found	npm install -g cli-jaw again. Check ~/.local/bin or npm bin -g is in $PATH
Error: node version	Upgrade to Node.js 22+: nvm install 22
NODE_MODULE_VERSION mismatch	npm run ensure:native (auto-rebuilds native modules)
EADDRINUSE: port 3457	Another instance running. Use --port 3458 or stop it first
Telegram / Discord auth fails	Run jaw doctor, check tokens, restart jaw serve
Browser commands fail	Install Chrome/Chromium. Run jaw browser start first
Employee dispatch hangs	Ensure the employee CLI is authenticated (jaw doctor)
Computer Use not working	macOS only. Codex CLI required. Check Automation permission in System Settings

---

Contributing

1. Fork and branch from master
2. npm run build && npm test
3. Submit a PR

Bug reports and feature ideas: Open an issue

---

MIT License · Built by developers who got tired of tab-switching between AI apps.