A lead hands out the work
An arrow from a lead to its helpers lets it send them tasks and wait for their results, on any mix of CLIs.
Learn moreNewnsq — NeuroSquad in your terminal. Free and open sourceNewvibereceipt — print a store receipt of your AI coding
New: Antigravity CLI and OpenCode 2
NeuroSquad — Run several coding agents.Stop babysitting them.
NeuroSquad calls you the moment an agent needs you. Claude Code, Codex, Antigravity CLI and 19 more on one screen, with your own logins.
Free · Windows 10 and 11 · macOS 12+ on Apple silicon and Intel
A replica of the NeuroSquad window. Four workspaces run at once: storefront on this computer, payments-api on a server over SSH, ml-eval inside WSL, and tiny-db with agents on local models. Each has its own squad of agent CLIs — Claude Code, Codex CLI, Cursor CLI, Qwen Code, OpenCode, aider, Goose, Kilo Code, Antigravity CLI and GitHub Copilot CLI — as live terminal cards next to a browser, terminals, a note, a task board and a budget card. Plugins (RTK-AI Token Saver, Caveman, Memory, Code Graph, Context7 docs), a skill and an MCP server are connected to agents by arrows, and the agents call their tools. Each agent shows whether it is working, waiting for you or done, and a notification says so. You can switch workspaces, open a card, switch between Canvas and Grid, draw or remove an arrow, approve a request, and pause the tour.
storefront
NeuroSquad · DoneUI · storefrontThe total is $72.25 with SAVE15 — correct.Done and Needs input are different signals — answer from the card or the notification.
Works with the agents you already use
Running agents shouldn’t mean watching them.
You flip between terminals every two minutes to see who is done.
An agent waited twenty minutes for a “yes” you never saw.
You walk back to your desk just to type “y”.
NeuroSquad watches every agent and calls you when one needs you.
Never miss a waiting agent
It calls you the moment an agent needs you
Status comes from each CLI’s own hooks, not from guessing at terminal output. Working, needs you, finished: a sound, a notification with the actual question, and an answer from wherever you are.
- Exact status from the CLI
- Claude Code, Codex and the others report every turn through their own hooks, so “Needs you” means a real question is waiting.
- The question in the notification
- The notification shows the command or question itself, so you know whether it can wait before you open the card.
- Answer from your phone
- Pair your phone with a QR code, then answer, type a prompt or stop an agent over Wi-Fi or a tunnel.
- Every agent on one screen
- A canvas or a grid with every terminal live, and a Squad panel that puts whoever needs you first.
shop-apiSquad status
0
Needs you
4
Working
0
Finished
- migrationsAdd the refunds tableWorking
- checkout-testsCover checkout end to endWorking
- docsDocument the refunds APIWorking
- reviewReview the refunds branchWorking
9:41
shop-apimigrationsWorking
● Write(db/migrations/0042_refunds.sql)
⎿ Wrote 18 lines
● Bash(npm run migrate -- --env staging)
⎿ 3 migrations applied
· Updating the refunds model…
Tap “1. Yes” on the phone, or just watch.
Squad at scale
Dozens of AI agents on one screen
Every card shows where its agent stands — working, finished or waiting for you. Zoom out to see the whole squad, click the one that needs you, and the camera takes you straight there.
Click a card and the camera flies to it; the fit-view button zooms back out.
Any CLI, side by side
Claude Code, Codex, Antigravity CLI, GitHub Copilot, Cursor, OpenCode and 16 more — every agent is a live terminal card on the same canvas.
Finished and Needs you are different signals
A green Done when an agent ends its turn, an amber Needs input when it stops to ask — and the notification says what it is asking.
One click to the agent that’s waiting
Click a card, a row on the Squad status card or a notification: the camera flies to that agent and its card flashes.
Frames keep big squads tidy
Group agents into frames by project or role. Zoomed out, each card becomes a tile with its main fact; zoom in to work.
Any CLI, any model
Not locked to one CLI or model
Mix Claude Code, Codex, Antigravity CLI, OpenCode and 18 more in one squad. Each one runs as itself, with your own login, and you can change your mind mid-session.
22 agent CLIs, each with its own page
Switch the CLI or model mid-session
Move a card from Claude Code to OpenCode or Codex. It keeps its arrows and notes, and the conversation comes along as the first message, taken from the transcript rather than summarised by a model.
A reviewer on another CLI
Each agent can work in its own git worktree. “Add a reviewer” puts a different CLI on the same tree, with an arrow from the author and a prompt not to edit anything.
Your models, your accounts
OpenRouter models, or your own server: LM Studio, Ollama, llama.cpp, vLLM. Keep Personal and Work logins for Claude Code and Codex and continue on the other one when a limit hits.
- OpenRouter
- LM Studio
- Ollama
- llama.cpp
- vLLM
UsageThis week, shop-apiSample numbers
- migrationsOpus 5.51,284,516 $4.62
- checkout-testsgpt-5.5846,203 $1.95
- docsAntigravity CLI312,980 $0.53
- reviewqwen3-coder-30b · LM Studio205,114 no price
- Total2,648,813 $7.10 + no price
Claude Code plan
5-hour window58%
Weekly34%
Workspace budget3,200,000
Agents running
Usage
Know what every agent costs
Tokens and cost per card, read from the CLIs’ own transcripts and counted to the token. A model without a known price says “no price”, never $0.
- Per card, per hour, per workspaceThe Usage page and the Run Stats card split the spend by agent, subagents included.
- Plan limits in viewThe 5-hour and weekly windows of Claude Code and Codex, from what the CLIs report. Your login is never read.
- A budget brakeAt the workspace limit, agents get Ctrl+C and automated prompts stop. Sessions are kept: raise the limit and work goes on.
Arrows are tools
Cards that talk to each other
Draw an arrow from an agent to any card and the agent gets that card’s tools through NeuroSquad’s built-in MCP server. No config files, no plugins — the canvas is the wiring diagram.
Tap a card to see what the agent can do with it
Browser
A real Chrome the agent navigates, clicks, types into, reads and screenshots.
browser_navigatebrowser_clickbrowser_typebrowser_snapshotbrowser_screenshot
Squads
Squads, when you want more
One agent is fine. When you want a team, an arrow decides who can use what, and everything else is one card away.
Looking for your own job to hand over?
The catalog has every scenario on one page: search by task, by card or by field — CRM, Telegram channels, marketplaces, code.
Browse all 40 scenariosDesktop or terminal
Two ways to run your squad
The same promise in two places: you run several coding agents and only step in when one needs you. Pick by where you spend your day, or use both.
NeuroSquad desktopThe main app
For when you want to see the whole squad at once and give agents tools, not just prompts.
- 22 agent CLIs as live terminal cards on a canvas, or in a grid
- Arrows between agents: hand out work, share MCP servers, skills and plugins
- Signals with the question itself, on the desktop, your phone or Telegram
- Usage per agent, workspace recording, templates and canvas mode
Download NeuroSquad
Free · Windows and macOS · NeuroSquad account
nsqFree, open source (MIT)
For people who live in the terminal, and for servers you reach over SSH.
- A keyboard-first grid of live terminals: Claude Code, Codex, OpenCode or any command
- A daemon keeps agents working after you close the terminal, and nsq updates itself while they are idle
- “Needs you” from each CLI’s own hooks, with the question; answer with y, a or n
- OpenRouter or your own model servers, phone answers (experimental), no account, no telemetry
Both are built on one open integration core, so status works the same way in each.
Trust
Read the privacy noticeYour CLIs, your logins, your machine
NeuroSquad starts the agent CLIs you installed, signed in with your own subscriptions and keys, and writes nothing to their config files. Your code, prompts and terminals stay on your computer. A free NeuroSquad account lets us count how the app is used: numbers only, never code, files, prompts, names or paths.
Your own CLIs and logins
We launch the CLIs you installed. Your subscriptions and API keys stay with them, and NeuroSquad never reads them.
Their configs stay untouched
Hooks, MCP and settings are added per card through flags and folders of its own, never written into your CLI’s config.
Your work stays local
Workspaces, notes and history are JSON files on your disk. Removing a card never touches your project folder.
Counts, not content
Usage statistics are numbers: online status, installs, which agents and cards are used, how big a workspace is.
Windows and macOS
One line to install. It keeps itself up to date.
The same free app on both systems. Paste a command or take the installer — after that, NeuroSquad updates itself.
Windows
Windows 10 or 11 · 64-bitwinget
winget install NeuroSquad.NeuroSquadScoop
scoop bucket add neurosquad https://github.com/glmn-ai/scoop-neurosquadscoop install neurosquadPowerShell
irm https://neurosquad.ai/install.ps1 | iexOr the 160 KB web installer from the download page.
macOS
macOS 12 or newer · Apple silicon and IntelTerminal
curl -fsSL https://neurosquad.ai/install.sh | bashHomebrew
brew install --cask glmn-ai/neurosquad/neurosquadOr a .dmg for your chip from the download page.
Workspaces where your code lives
A workspace is a project folder — and it doesn’t have to be on this computer.
- This computerEvery agent CLI, every card.
- WSLA Linux folder in a WSL distro on Windows. Shells and Claude Code run there today.
- SSHA folder on a remote host, over one shared, verified connection. Shells and Claude Code run there today.
Questions
Stop babysitting your agents
Free for Windows and macOS. Bring the CLIs you already use.
Download NeuroSquadWindows 10 / 11 · macOS 12 or newer, Apple silicon and Intel
Linux is coming later.
The first-run wizard can install an agent CLI, Git and the dictation model for you — or skip each step.
