SquadraHQ Español

NewAn inbox for your decisions, and a plan for every teamAn inbox for your decisions

Your AI agents, run as a team.

A coordinator splits the work among its workers, on any of your machines, and approves what is safe. What is yours to decide waits in your inbox, so you can be somewhere else.

Get started View on GitHub

Open source, MIT. Self-hosted: one Node.js server, no dependencies.

Works withClaude CodeCodex CLI experimentalGrok CLI experimental
The SquadraHQ inbox: a permission waiting and two decisions from the Development coordinator, one open with its options and the recommended one marked

The inbox

Only what is yours to decide.

Coordinators leave you one question per decision, with its options and what they recommend. Everything else they settle themselves, and keep going while they wait.

  • Answer in one tap. Pick an option, add a detail if you want, from your desk or your phone.
  • Never on old information. When a coordinator learns something new, it updates the question; you see it changed.
  • Permissions too. A worker waiting to run a command shows up here, with exactly what it wants to run.
  • Or put it away. Dismiss a question and its coordinator is told not to wait for an answer.
A decision on a phone: the question, two options with the recommended one marked, and room to answer in your own words

Teams

Every team, with its plan in sight.

A team is a coordinator and its workers, each in its own repository and with its own persona. The coordinator keeps its plan up to date: what is done, what is in progress, what is blocked and what waits on you.

  • How far it is, at a glance. Each task with who has it, its state and where it comes from.
  • Linked teams. Design and marketing, say, hand work to each other through their coordinators.
  • Templates. Save a team and open the same one again, on any machine, in one go.
A team's page: its plan with three of seven tasks done, one in progress, one blocked and one waiting on you, and its workers' latest reports

Live

Watch them all at once.

Every agent is a normal interactive Claude Code session: see its conversation, its tools and its terminal live, and step in and type whenever you want.

A team all at once: the coordinator's terminal and its workers' conversations side by side

Each agent

What it did, without reading it all.

Beside each agent’s terminal, what it is doing now and a timeline of what mattered. The terminal stays exactly as it is.

  • Now, at a glance. Its state, how full its context is and its task in the plan.
  • What mattered. Its orders, the files it edited, tests and commits, permissions and its reports, newest first.
  • No tokens spent. Read from what each session already reports: nothing is asked of the model.
A worker’s terminal and beside it what it is doing now and its timeline: a report, tests passed, a file edited and the order it was given

Everything else

Built for leaving it working.

All your machines

Each agent runs on any computer you add, with that computer's own login: Claude Code, or Codex CLI and Grok CLI (experimental). Machines connect out: no open ports.

Safe by default

The coordinator approves only what has no side effects; anything destructive, irreversible or costly comes to you first.

Ctrl+K anywhere

Go to any session, team or decision, or open an action, from the keyboard.

Companies and people

Several companies on one server, each apart; accounts with codes and a role each.

Your own skills

Give each agent a skill or agent from its project as its persona: design, marketing, code review…

Self-hosted

One Node.js server or a Docker image, automatic HTTPS when you give it a domain. MIT licensed.

Get started

Up and running in three steps.

Start the dashboard

On any computer with Node.js 22, or with Docker. The first time, its token makes you the owner.

Add your machines

Machines → Add machine walks you through it: one command for each computer where agents will work, and it tells you when it has connected.

Open a team

New team: what it is for, who is in it and where it runs. Then give its coordinator its first order.

git clone https://github.com/roke22/SquadraHQ.git squadra && cd squadra
git checkout "$(git tag --sort=-v:refname | head -1)"   # the latest release
node server/server.js                                    # http://127.0.0.1:7787

Docker, HTTPS on a domain and everything else: the README and the reference.

How it compares

More than one terminal.

WhatClaude Code subagentsClaude Code Agent TeamsSquadraHQ
Agents run onthe same sessionthe same machineany machine you add, each with its own login
You follow themin one terminalin one terminal eachin one live dashboard, from any browser or phone
Your decisionsasked in the terminalasked in the terminalin your inbox, one per decision, with options
Unattended workyou approve each promptyou approve each promptthe coordinator approves what is safe
Several repositoriesnothe lead's folderone worker per repository, where it lives

What you need

Claude Code, and a computer.

A Claude Code accountAny claude.ai plan with Claude Code (Pro, Max, Team or Enterprise) or an Anthropic Console API key.
One computer for the dashboardWith Node.js 22 or Docker. It can be the same one where your agents run: Linux, macOS, or Windows with WSL.

Any coordinator or worker can also run Codex CLI or Grok CLI, with that agent's own account: experimental, see how it differs.

Put your first team to work.

Get startedSee examples