agenticsquad.dev · v1.3.0

Talk to one routine. The team finishes it.

You describe the work once, to the Scrum Master. It sizes the request and puts it on a Kanban board as cards. The Tech Lead, Developer and QA pick those cards up and finish them, and you watch each handoff in a live office and on the board.

See a request move

First cd into your project folder, then run npx agenticsquad. Setup takes under 5 seconds.

Why

You should not have to run a team of agents by hand. Talk to one routine and the team finishes the task.

Your session hands the request to the Scrum Master and only dispatches from there. It picks the platform team (Web, Android or iOS). A small change goes straight to that team's Developer with its Tech Lead reviewing; a bigger one goes to the Tech Lead, who writes the contract and brings in QA. Instructions, decisions and results live on the cards, so no one works from a pasted prompt.

Usually all of that is a scrolling terminal. AgenticSquad.dev reads the sessions your tools already record and shows who is working, on which card, what they handed off and why.

How one request moves

How the AI agent team turns a sentence into a finished card.

Small change? Fast path.

When the Scrum Master sizes a request as small (one to three files, no change to a shared type, API, payload or stored data), it skips the contract. One card goes straight to the platform's Developer (web-dev, android-dev or ios-dev), who maps the impact first, and that platform's Tech Lead reviews it.

youYou Scrum Master Developer Tech Lead reviews
What you see

Every angle on your AI agents' work, all from real sessions.

Next to the floor plan and the board, the portal shows the conversation between agents, the tools they used, the code they changed and the time and tokens each role spent. A finished requirement can be replayed on the floor plan, and first-time users can watch a short demo.

Handoff stream

Every delegation with the exact brief that was passed, so you can see why a task moved.

Example conversation

Tool analytics

Which tools each agent reached for: terminal commands, MCP tools, handbacks and file writes, counted per call.

Tool calls the portal recorded, all agents, measured 17 Oct 2026. Shell commands logged by name (cd, curl, ls…) are counted as Bash; every mcp__ tool is counted as MCP; write_to_file as Write.

Code change inspector

The diff each agent produced, filed under the card and the role that made it.

src/components/KanbanBoard.tsx+3 −1
Example diff · web-dev · VO-43

Time and tokens

Work time is logged per role while it is active, with the token usage your tools recorded.

Example figures, counting up

Replay a requirement

A Replay button on a finished requirement plays its handoffs on the floor plan, with decisions, reviews and results as captions. Save as video keeps a recording. Nothing is written to the board.

Example timeline · VO-41

Watch demo

New to the office? A Watch demo button on the floor plan walks a scripted request through your own team, from the Scrum Master to the people who build and test it and back. It runs in the browser and saves nothing.

Built from the roster you saved in Team Studio
Works with your AI tool

Define the agent team once. Claude Code, Codex, Gemini and OpenCode get the same roles.

Roles, skills and rules live as plain Markdown and YAML in your repository, and Team Studio in the portal edits the same files. One command writes them out in the format each AI tool reads, including the rule that the main session only dispatches, so nothing drifts between tools.

Your team, in the repo

.agents/registry.yaml
.agents/definitions/tech-pm.md
.agents/definitions/web-tech-lead.md
.agents/definitions/web-dev.md
.agents/definitions/web-tester.md
.agents/definitions/android-tech-lead.md
.agents/definitions/android-dev.md
.agents/definitions/android-tester.md
.agents/definitions/ios-tech-lead.md
.agents/definitions/ios-dev.md
.agents/definitions/ios-tester.md
.agents/skills/ · 60 skills
npx -p agenticsquad agents sync
Claude CodeCLAUDE.md · .claude/agents/
OpenAI CodexAGENTS.md · .codex/
Gemini · AntigravityGEMINI.md
OpenCode.opencode/agent/
Activity is read fromClaude Code transcriptsOpenCode databaseAntigravity sessionsNothing to install in the agent tool.
Safe to run on a work laptop

It stays on your machine and shows only what really happened.

Local only

The server listens on 127.0.0.1 and refuses requests from other websites, so a page you visit cannot read your board.

Tokens stay private

Jira and Confluence tokens are saved to a gitignored file readable only by you. The portal never displays them.

No made-up numbers

There is no mock data. Every card, handoff and token count comes from a real session, and 33 test suites guard it in CI. The Watch demo is labelled as a simulation and never touches your board.

Changes go through the team

The session you type into only dispatches work. In Claude Code a hook stops it from editing files, so every change, even a one-line fix, goes to the Scrum Master and gets reviewed.

Get started

Install the agent team with one command.

  1. 1

    Open your project folder in a terminal. New or existing, any stack.

  2. 2

    Run the command from inside that folder: cd your-project, then npx agenticsquad. It takes under 5 seconds. The portal finds your agents, or helps you pick an Android, iOS or Web team (Scrum Master, Tech Lead, Developer and QA), or all three with the Scrum Master routing by platform, that you can adjust in Team Studio.

  3. 3

    Ask as usual in Claude Code, Codex, Gemini or OpenCode. Your session hands the request to the Scrum Master, and the office fills in as your agents work.

~/your-project
$ cd your-project
$ npx agenticsquad
Discovered 4 agents in .agents/registry.yaml
SM · LEAD · DEV · QA
✓ Portal running at http://127.0.0.1:3000
Watching Claude Code, OpenCode and Antigravity sessions
FAQ

Questions about the AI agent team.

New to it? Read the guide to running a multi-agent team in Claude Code.

What is AgenticSquad.dev?

AgenticSquad.dev gives your AI coding assistant a team: a Scrum Master plus a Tech Lead, Developer and QA for each platform (Web, Android, iOS). You talk to the Scrum Master, it puts the work on a Kanban board as cards, and the other agents pick the cards up and finish them. A local portal shows every handoff live on an office floor plan and on the board.

How do I install it?

You need Node.js. Open a terminal, cd into your project folder, then run npx agenticsquad. There is nothing to install globally, and it takes under 5 seconds. The portal opens in your browser and finds your agents, or helps you pick a Web, Android or iOS team.

Which AI coding tools does it work with?

Claude Code, OpenAI Codex, Google Gemini and Antigravity, and OpenCode, including local models through OpenCode. You define the team once and npx -p agenticsquad agents sync writes it out as CLAUDE.md, AGENTS.md, GEMINI.md and each tool's own agent files.

Does my code leave my machine?

No. The portal runs on 127.0.0.1 and reads each tool's session history from your own disk. Code, file paths, card text and tokens are never sent anywhere. The portal sends anonymous usage statistics, which you can switch off in Settings or by starting it with VO_ANALYTICS=off.

Is it free?

Yes. AgenticSquad.dev is free and MIT licensed.

Can I change the agents, their skills and rules?

Yes. Edit them in Team Studio inside the portal, or edit the Markdown files in .agents/definitions/ directly. You can set roles, names, allowed tools, skills, rules and avatars, then run npx -p agenticsquad agents sync.

How do small changes differ from big ones?

The Scrum Master sizes each request. A small change (one to three files, no change to a shared type, API or stored data) goes straight to the platform's Developer, with the Tech Lead as reviewer. A bigger one goes to the Tech Lead, who writes technical contracts for the Developer and hands testing to QA.

Can I run it on another port or for another folder?

Yes. Use npx agenticsquad --port 3001 for another port, or npx agenticsquad --dir /path/to/project to watch a different project. Several projects can run at once on different ports.