# Poly vs Nimbalyst for teams using Claude Code and Codex

Nimbalyst shares what the agents write. Poly shares the agent.

Published 2026-10-05 by Richard Kaminsky and Mitchell Lipyansky, the co-founders of Poly. Canonical: https://usepoly.co/poly-vs-nimbalyst
Poly is a multiplayer AI coding workspace: a shared room where your team works with one AI agent, together. Free to start: https://usepoly.co/

**Nimbalyst and Poly both call themselves multiplayer for coding agents, and they mean different things. Nimbalyst is an open-source (MIT) desktop workspace for Mac, Windows and Linux, with an iOS app, that wraps your own Claude Code, Codex, OpenCode and Copilot agents in visual editors for Markdown, mockups, diagrams, spreadsheets and trackers. Its Teams plan shares those documents and trackers live, while each person's agents keep running on their own machine. Poly (usepoly.co) shares the agent itself: a browser room where everyone watches the same session, any member can approve a change before it lands, and every turn ends with a receipt.**

## Key takeaways

- Nimbalyst: each person runs their own agents locally, and the team shares documents, mockups and trackers in real time.
- Poly: the team shares one agent session in a browser, with nothing to install and no keys.
- Nimbalyst is open source and free for individuals; Teams is $20 per user a month and free during its beta. Poly is free to start, then $20 or $50 per person who runs the AI, models included.

*Poly and Nimbalyst, from Nimbalyst's site, pricing and changelog, October 5, 2026.*

|  | Poly | Nimbalyst |
| --- | --- | --- |
| What the team shares | One live agent session | Documents, mockups and trackers |
| Where agents run | A sandbox on Poly's servers | Each person's machine, or their own Cloudflare sandbox |
| Agents | Claude Code and Codex, run by Poly | Claude Code, Codex, OpenCode and Copilot |
| Models | Claude, GPT, Grok, Kimi or Muse, included | Your own subscription, key or local model |
| Approve AI changes | Any member, before each change, in the modes that ask | Each person approves their own agent's diffs |
| Open source | No | Yes, MIT (desktop and iOS apps) |
| Platforms | Any browser | macOS, Windows, Linux and iOS |
| Price | Free start; $20 or $50 per person who runs the AI | Free; Teams $20 per user a month, free in beta |

## What is Nimbalyst?

[Nimbalyst](https://nimbalyst.com) is a visual workspace for coding agents. It runs parallel Claude Code and Codex sessions, optionally in separate git worktrees, next to editors for Markdown, CSV, mockups, Excalidraw, Mermaid, data models and more, and it shows every AI edit as a diff you approve before it lands in the file. The desktop and iOS apps are MIT licensed, with about 1,800 stars on GitHub. You bring your own AI access: a subscription, an API key, or local models through LM Studio.

[Nimbalyst Teams](https://nimbalyst.com/teams/how-it-works/) adds a shared layer. In Nimbalyst's words, everyone keeps working locally with their own agents, and the documents and trackers you choose to share become a layer the whole team works on together, in real time. A teammate who doesn't run the app can open a shared plan in the browser, edit it and comment. Crew, in alpha and off by default since September 29, adds persistent agent teammates that work scheduled shifts within token budgets.

## How does a team work together in each?

**Nimbalyst.** Each person prompts their own agents. When something should be shared, such as a plan, a mockup or a tracker, they promote it to a shared file, and teammates and their agents edit it together, with its history kept. Nimbalyst's docs describe shared documents rather than a shared agent session.

**Poly.** The session is the shared thing. Everyone in the room sees each prompt, reply and change as it happens, anyone can prompt, and in the modes that ask each proposed change is a plain-language card that any member can approve or reject. A group chat sits beside the timeline where the AI can't read it, and every turn is a commit anyone can undo. For separate work, the room's owner turns on multi-agent mode, which gives each member their own agent on their own branch.

## Where Nimbalyst is stronger

- **Open source,** under MIT, on every desktop system.
- **Visual editors** for mockups, diagrams, data models and spreadsheets, in the same app as the agents.
- **Your own agents and models,** including local ones.
- **Shared plans and trackers** that people and agents edit live, readable in a browser by people who don't install it.
- **Diff approval** on every AI edit before it reaches a file.

## Where Poly is stronger

- **One live session for the team.** Everyone sees the same turn as it runs and can redirect it.
- **Approvals from any member,** before the change.
- **Nothing to install and no keys.** The models are included.
- **A receipt for every turn,** naming who ran it and what it cost.
- **A chat the AI can't read,** and watch links for anyone without an account.
- **A hosted sandbox,** so the agent doesn't run on anyone's laptop.

## Pricing, side by side

[Nimbalyst](https://nimbalyst.com/pricing/) is free for individuals, with no account or card. Teams, in beta, is $20 per user a month or $200 a year, and free during the beta. Enterprise runs the collaboration server in your own Cloudflare account or has Nimbalyst manage it. Model usage is on your own accounts. Poly is free to start with no card; each person who runs the AI pays $20 for Pro or $50 for Max, models included, and watching, chatting and approving are free.

## Which should you pick?

- **Each engineer wants their own agents locally, and the team needs shared specs, mockups and trackers:** Nimbalyst.
- **You want open source, or you're on Windows or Linux and want a desktop app:** Nimbalyst.
- **Several people, including some who don't code, need to steer and approve the same agent live:** Poly.
- **You want the models included and a receipt for every turn:** Poly.

Nimbalyst also publishes a comparison with Conductor; ours is [Poly vs Conductor](/poly-vs-conductor). More tools are in [Collaborative Claude Code tools](/collaborative-claude-code-tools).

## Common questions

**Is Nimbalyst multiplayer?**

Its site says it is multiplayer for Claude Code and Codex. What its Teams plan shares is documents, mockups and trackers, edited live by people and agents, while each person's agents run on their own machine.

**Is Nimbalyst free?**

Yes for individuals, with no account needed, and the apps are MIT licensed. Teams is $20 per user a month and free during its beta. You pay your own AI provider for model usage.

**Does Nimbalyst work on Windows?**

Yes. Nimbalyst has desktop apps for macOS, Windows and Linux, plus an iOS app. Poly runs in a browser on any of them.

**Can teammates steer one agent session in Nimbalyst?**

Nimbalyst's docs describe shared documents and trackers with each person's agents running locally, not a shared agent session. In Poly, several people steer one session together.

More guides: https://usepoly.co/guides · Security: https://usepoly.co/security · Pricing: https://usepoly.co/pricing