Skip to main content
ManyAgentsDownload

Many Agents.One Window.

Run a team of Claude Code and Codex agents on your Mac. Hand them a project, watch the work take shape beside the chat, and step in when a question lands.

Download for Mac
Buy for $49

$99 Early adopter price

macOS 14 or later · Claude Code sign-in · Version 1.4.6, Nov 10, 2026

ManyAgents window. Across the top, tabs for the Orchestrator, Cart rounding fix, Checkout flow, a Reviewer and Subscription pause and skip, three of them in their own git worktrees. On the left, the Orchestrator reports that the cart fix is verified on main and that a Reviewer running read-only on Codex is on checkout. On the right, taking half the window, the built-in browser shows the Brightside coffee storefront running locally, with three items in the cart.

The team

Instead of five terminal tabs, an orchestrator that runs them all.

Tell the orchestrator what you want. It splits the request into tasks, opens a tab for each and keeps track of who is doing what. You talk to it, and it talks to the rest.

The card overview for the brightside-coffee project. Its Orchestrator dispatches agents one at a time: workers for Cart rounding fix, Subscription pause and skip and Mobile nav overflow, each in its own worktree, and a thinker for Rate limiting design. The cart fix finishes and a Reviewer picks it up, then the thinker comes back marked needs you.

  • Worker

    Builds one task in its own git worktree, tests it and reports back. The orchestrator merges finished branches one at a time and stops cleanly on a conflict.

  • Thinker

    Takes the hard calls: a design decision, or a bug that beat two workers.

  • Reviewer

    Reads finished work and reports. It cannot edit, and it runs on Codex when that is set up.

Built-in browser

Agents test it in the window, and you watch.

Your app runs in a browser beside the chat. Agents open it, click, fill forms and read the result. Ask for a run through checkout on a new account, or every page at phone width. It stays signed in to your sites, and you can take over with a click.

An agent testing checkout in the built-in browser: it opens the store, adds a coffee, enters an email, places the order and reads the confirmation.

Needs you

Every question in one place, with a recommendation.

When an agent needs a decision, it lands here with what the agent would do. Click it to open the tab and answer, or mark it done.

The Needs you drawer beside a worker's conversation. It holds the worker's question about revoking refresh tokens with a recommendation, then the Orchestrator's question about shipping behind a flag arrives.

iPhone

Answer from your phone.

The iPhone app shows your projects and tabs, lets you answer questions and approve requests, and lets you message your agents. Your Mac keeps running them, so it stays awake and online.

Available through TestFlight. Request an invite.

Two iPhone screens. Projects lists plan usage and three projects, and api-gateway turns orange as a question arrives. Chat shows a screenshot of a checkout total sent to brightside-coffee, and the orchestrator's answer that the rounding fix is in and heading to review.

iOS Simulator

Run the Simulator beside your agents.

Flip the panel to iOS and use the build live while the agent works on it. Agents can tap through it and read the screen too. Each project keeps its own simulator.

ManyAgents window. On the left, an Orchestrator tab for ~/Sites/brightside-ios reports that the pause button bug is fixed and that a worker is building order history. On the right, the preview panel set to iOS shows the Brightside app running in an iPhone 16 Pro simulator.

The method

How I run my own board.

ManyAgents is built around one loop: I make a request, a thinker decides, workers build and a reviewer checks, until the work is accepted. It's opinionated, and every part of it is optional.

Read the method

Up and running in a few minutes.

  1. Download and open

    Your 14-day trial starts right away. No account, no card.

  2. Sign in to Claude Code

    The app walks you through it. Codex is optional.

  3. Open a project

    Tell the orchestrator what you want done.

What do I need?
A Mac on macOS 14 or later, and Claude Code signed in with a Claude Pro or Max subscription or an Anthropic API key. Codex is optional: with it, reviewers run on OpenAI models by default and any role can; without it, everything runs on Claude.
Does it cost extra to run agents?
ManyAgents uses the Claude and Codex access you already have, and shows how much of each limit is used. Your plan's limits and any API charges still apply.
What happens after the trial?
You can still read and export everything. Sending messages and launching agents pause until you add a license. A license is $49, one time, with 12 months of updates. See pricing.
What happens after 12 months of updates?
The version you have keeps working, and so does every version released during your 12 months. When updates end, Renew appears in the app and emails you a single-use code for 50% off the current price.
Is my code sent anywhere?
Agents run through Claude Code and Codex under your own accounts, so your prompts and code go to Anthropic or OpenAI the same way they would from the terminal. Nothing goes to us, and the app has no analytics. See the privacy policy.

Try it free for 14 days.

Download for Mac
Buy for $49

$99 Early adopter price

macOS 14 or later · Claude Code sign-in · Version 1.4.6, Nov 10, 2026