CLI-JAW

One assistant with shared memory and a dashboard on top of Claude Code, Codex, Cursor, Copilot and other AI CLIs you already have

  • 215 ★ on the repository
  • Last commit
  • MIT
CLI

Broad access

Why this level

  • The assistant runs agents with file and terminal rights and controls Chrome and the desktop, so a wrong instruction can change or erase data
  • Telegram, Discord and Slack tokens open remote control of an agent on your machine to anyone who gets access to the bot or channel
  • Running through vendor subscriptions may not match their terms, and the consequences, including account suspension, fall on the user
All reasons and checks
Needs a VPN

lidge-ai/cli-jaw

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add cli/cli-jaw

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • The assistant runs agents with file and terminal rights and controls Chrome and the desktop, so a wrong instruction can change or erase data
  • Telegram, Discord and Slack tokens open remote control of an agent on your machine to anyone who gets access to the bot or channel
  • Running through vendor subscriptions may not match their terms, and the consequences, including account suspension, fall on the user

This is third-party code. Review the repository files before installing.

What it does

CLI-JAW is the npm package `cli-jaw` with the `jaw` command. It joins the AI CLIs you have installed into one assistant. The main CLI (the "boss") hands pieces of work to others (the "employees"), each running on its own engine and model, and then reviews the result. Claude, Codex, Cursor, Grok, Kiro, OpenCode, Copilot, Antigravity and Pi are supported. Sign-in happens in the CLIs themselves with your subscriptions, for example `claude auth login`, and when one hits a limit the work moves to the next engine. For complex tasks there is the PABCD mode: plan, audit, build, check, done. A person approves every transition, and the verifying workers are read-only. State survives restarts. Memory has three layers: recent history, structured notes and search over them through `jaw memory search`. On top come over 200 ready skills (`jaw skill install <name>`) and one MCP setup: `jaw mcp install` syncs the server list into Claude, Codex, Kiro, OpenCode, Copilot and Antigravity at once. The package starts with `jaw dashboard` on port 24576, and the agent server `jaw serve` runs on port 3457. It has a web UI, a kanban board of sessions, Markdown notes, Chrome control through DevTools and Telegram, Discord and Slack channels. A desktop app is available for macOS, Windows and Linux.

Who it is for. People who pay for several AI CLIs at once and are tired of moving context between them by hand.

Good fit when

  • Frontend is easier to fix in OpenCode and backend in Codex, and you need one coordinator with a shared result
  • One engine hits its subscription limit and the work should move to the next one without manual switching
  • You want to give an agent tasks from your phone through Telegram, Discord or Slack
  • You need one MCP server list that syncs into several CLIs at once
  • A long task needs a plan with approval at every transition so the agent does not drift

Not a fit when

  • You use one CLI and do not plan to add others: a coordination layer adds needless complexity
  • You need a cloud platform with shared team access: CLI-JAW runs on your machine for one user per project
  • Your subscription or the vendor's terms do not allow running their CLI through a third-party orchestrator: check this before starting
  • You need a setup where the agent can do nothing on the machine: the assistant has access to the terminal and the browser

Example request

Give the dashboard layout fix to an OpenCode employee and the /api/users endpoint update to a Codex employee, then show a combined report

Limitations

  • It needs Node.js 22.4 or newer. The native Windows install is marked beta, and the recommended path there is WSL.
  • Signing in to the vendor CLIs remains your job, and whether your subscription supports such runs is set by each vendor's terms.
  • On npm 12 and newer the install can block dependency scripts, so install with the `--allow-scripts=cli-jaw` flag.
  • Desktop control through `jaw-computer-use` works on supported macOS and Windows hosts, and the Windows build is unsigned.

How to disable. Stop the server and the service with `jaw service stop`, then remove the package with `npm uninstall -g cli-jaw`. If you installed the desktop app, uninstall it the usual way. Revoke the Telegram, Discord and Slack tokens in the bots' own settings. The underlying CLIs stay in place.

FAQ

How do I install CLI-JAW?

Run `npm install -g cli-jaw`, then `jaw dashboard` and open `http://localhost:24576`. Node.js 22.4 or newer is required. For existing setups there is a gentle variant, `JAW_SAFE=1 npm install -g cli-jaw`, which skips optional tool setup.

Are API keys required?

Not necessarily: authentication goes through the CLIs themselves with your subscriptions or free tiers, for example `claude auth login`, `codex login` or `copilot login`. Cursor also supports `CURSOR_API_KEY`. Check everything at once with `jaw doctor`.

What does the PABCD mode do?

It leads a complex task through five phases: plan, audit, build, check, done. The boss writes the plan and stops for your approval, while read-only workers verify the plan and the result. Start it with `jaw orchestrate` or `/pabcd`.

How do I connect Telegram?

Create a bot with @BotFather and copy the token, then run `jaw init --telegram-token TOKEN` or set it in the web UI settings. Send the bot any message and the chat ID saves itself.

How do I add an MCP server to several CLIs at once?

The `jaw mcp install` command with a package name, for example `@anthropic/context7`, syncs the server into Claude, Codex, Kiro, OpenCode, Copilot and Antigravity in one step.

Discussion

No questions or reviews yet.

CLI tools

One runtime for Claude Code, Codex, OpenCode and more: a run counts as done only when checks the agent did not write pass

CLIBroad access606Repository stars

IJFW adds shared memory, plan gates and cross-audit to Claude Code, Codex, Gemini CLI, Cursor and other coding tools

PluginExtended access215Repository stars

A CLI that connects local agents to a shared cloud store of sessions, memory, skills and MCP connectors.

CLIExtended access107Repository stars

Self-hosted system for a team of AI agents: org chart, tasks, budgets and oversight of Claude Code, Codex and OpenClaw

CLIBroad access113.3KRepository stars

DevChain runs a team of AI coding agents on your own machine with a shared board, chat, code review and live terminals.

CLIBroad access135Repository stars

ClawMetry reads the session files agents already write and shows the timeline, tool calls, tokens and cost in one dashboard.

CLIExtended access436Repository stars

Comparison with similar tools

Cards that share the main search query, and how they differ.

ToolOfficialRepository starsMain difference
CLI-JAW (this card)no215
Claude Squadno8,790runs several agent sessions in isolated working copies, while CLI-JAW splits one task across different CLIs with shared memory, PABCD and messengers
CLI Agent Orchestratorno1,548an orchestrator for a set of CLI agents, while CLI-JAW adds a dashboard, MCP sync and control from Telegram, Discord and Slack

README in short

The page shows installation through npm and the desktop app, ways to sign in to the supported CLIs and a table of engines with default models. It then describes the boss-and-employees mode, PABCD, memory, skills, MCP, browser automation, messengers and a reference of `jaw` commands.

README badge

Report a problem