Free · Open source · MIT license

One chat.
Three minds.

Claude, Gemini and Codex in a single desktop app. Pick a model and talk. When a second opinion would change the answer, it asks the others.

ZIP · Windows 10 and 11 · needs Node.js 20 or newer

The Big Three app: Claude asks Gemini a question and shows the answer inside the conversation

The idea

You don't pick a workflow. You pick a model and talk. It reads, plans and edits on its own, and when an independent review would make the result better, it brings in Claude, Gemini or Codex. Every consultation is right there in the chat.

Features

Everything the three can do,
in one calm window.

A team that consults itself

The model you talk to decides when a colleague should weigh in, or you just ask: "ask Gemini to review this". Each question and answer appears as a card in the conversation.

Real model choice

Opus, Sonnet and Haiku with an effort level. Every model your Antigravity account offers. Any Codex model by name. Each teammate can answer on its own model.

  • Opus
  • Sonnet
  • Haiku
  • Gemini Flash
  • Codex

Safe by default

Models start read-only. Letting one edit files or run commands asks you in a native dialog, every time.

Your database, never your password

Connect PostgreSQL, MySQL or Oracle. Models run SELECT queries through a broker; the password stays in the Windows vault.

It remembers

The lead model keeps its own session between messages, so it remembers what it read and did, just like in its own CLI.

Your subscriptions, no API keys

Big Three drives the official CLIs on your own accounts. Each keeps its own sign-in and its own limits.

Accessible and light

Light, dark or automatic theme, four text sizes, reduced motion, keyboard navigation and screen reader labels.

How it works

Three steps. Then just talk.

  1. 1

    Pick who you talk to

    Claude, Gemini or Codex, with the model and effort you want. Start a conversation in the folder they may work in.

    Start screen of Big Three
  2. 2

    Choose the team

    Switch colleagues on or off and give each one its own model. They read and answer; only the lead changes files.

    Choosing which colleagues can be consulted, and on which model
  3. 3

    Ask anything

    The lead works and, when it helps, consults a colleague. You see every question and answer as it happens.

    Claude consulting Gemini during a conversation

Safety

Read-only until you say otherwise.

Big Three lets AI act on your computer and query your data, so its rules are strict and visible.

You confirm every write
Editing files or running commands needs a native confirmation the page itself cannot answer. It is not remembered between sessions.
Colleagues cannot change anything
A consulted model runs read-only, without room or database tools, so questions cannot fan out.
Defence in depth for databases
A SELECT-only user, a SQL guard, read-only transactions, timeouts, row limits and an audit log.
Nothing sent anywhere
Big Three itself has no telemetry. The CLIs talk to their providers exactly as they normally do.
Connecting a database: PostgreSQL, MySQL or Oracle

Download

Ready in a few minutes.

Free and open source. Everything runs on your machine, with your accounts.

You need

  • Windows 10 or 11
  • Node.js 20 or newer · nodejs.org
  • At least one CLI signed in: Claude Code, Antigravity (Gemini) or Codex
  1. Download the ZIP and unzip it.
  2. Double-click Big-Three.cmd. The first run installs what it needs.
  3. Sign in to any model from the sidebar, and start talking.
Prefer the terminal?
git clone https://github.com/marcj19/Big-Three.git
cd Big-Three
npm install
npm start

Run npm run doctor to see which CLIs are installed and signed in.

FAQ

Questions, answered.

Is Big Three free?

Yes. It is free and open source under the MIT license. You only use the subscriptions you already have for Claude, Gemini and Codex.

Do I need API keys?

No. Big Three drives the official command-line tools signed in with your own accounts, so it uses your subscriptions and their limits, never pay-per-token API keys.

Can the models really talk to each other?

The model you talk to can consult Claude or Codex as a lead, and any of the three as a colleague, through a tool built into the app. Each consultation appears in the conversation.

Does the AI see my database password?

No. The password stays in the Windows Credential Manager and is read only by the database broker. Models can only send SELECT queries, and you should give them a database user that can only read.

Does it work on macOS or Linux?

It is built and used on Windows 11. The automated tests also pass on Linux, but the app itself has not been tried on macOS or Linux yet.

Is it made by Anthropic, Google or OpenAI?

No. Big Three is an independent, unofficial project and is not affiliated with or endorsed by Anthropic, Google or OpenAI.

What language is the app in?

The interface is in Brazilian Portuguese for now. The models answer in the language you write in, and translations are welcome on GitHub.