Why I Need Both Codex and Claude Code: Introducing the Tutti Agent Operating System

Hello, I’m Zhang Beihai from AI Learning.

Problem: Agent information silos

When creating text‑image or video content, I use Claude Code for text generation and Codex’s Image2 for image creation; for video, Codex handles code execution. Switching between agents requires manual copy‑paste of files, context, and commands, similar to using Claude Code for design proposals and Codex for implementation.

Solution: Tutti’s unified workspace

Tutti pulls all agents into a single real‑time workspace, effectively acting as an “Agent operating system” with a file system, applications, browser and terminal.

Claude Code and Codex share the same files, context and running tasks, so Codex can see Claude’s edits instantly without manual sync.

The platform also includes an App ecosystem (image generation, UI/UX design, documentation, PPT, etc.) that agents can invoke directly; for example, Codex can call an image‑generation App and Claude Code can immediately use the result in front‑end code, eliminating copy‑paste.

Key feature: @ reference

Typing @ in a chat lets you embed any resource—unfinished dialogue, long document, an App from the market, or a half‑finished task—into the current context without downloading, uploading, or re‑explaining.

1. Cross‑Agent usage

In Claude Code, @ a local file makes it instantly readable; @ a previous conversation restores its context.

Tasks and apps can also be pulled on demand.

2. Agent switching

After Claude writes a back‑end API, switching to Codex with @ brings the dialogue and API docs so Codex can immediately start front‑end work. Codex can run tests in parallel while Claude continues other tasks.

App ecosystem

Tutti’s App market offers three usage modes:

Direct use : Install an App (e.g., AI Canvas) and generate images on a canvas via prompts.

@ invocation : In an agent conversation, @ an App (e.g., image generation) and let the agent handle the result, then @ Codex to insert the image into code.

Task‑center orchestration : Break complex work into sub‑tasks, assign each to a specific App or agent.

All Apps reuse your existing Claude Pro or Codex subscriptions at no extra cost.

Control center

A single view shows every agent’s dialogue, pending approvals, and running tasks, allowing one‑click navigation and approval.

Who should use Tutti

Independent developers : Claude proposes, Codex implements without re‑explaining the project.

Designers : Use design Apps to create drafts, then let Codex turn them into code directly.

Product managers : Codex writes PRDs, then automatically calls UI/UX Apps to generate prototypes without opening Figma.

Any workflow involving two or more AI agents benefits from Tutti’s reduction of “hand‑off” overhead.

Installation and usage

The desktop client is currently macOS‑only; Windows users can build from source.

# Prerequisites: Node.js 24+, pnpm 10.11.0, Go 1.24
pnpm install
pnpm setup:dev
make dev-gui

My assessment

Tutti finally addresses the long‑standing need for lightweight multi‑agent collaboration. Existing solutions are either heavyweight (requiring extensive orchestration code) or simplistic (clipboard‑only).

The design does not replace any agent; it merely connects them and provides a GUI workbench.

Pros

The @ design elegantly solves ~90% of context‑transfer friction.

App reuse of existing subscriptions is user‑friendly.

Open‑source and local‑first keep data private.

Pure GUI lowers the barrier for non‑technical users.

Cons

Early Access – Windows client not yet released.

VM version unavailable, limiting multi‑user collaboration.

App ecosystem still small; users may need to create their own Apps.

Building from source requires Node.js 24+ and Go 1.24, which raises the entry barrier.

Overall, the direction and execution are solid. If you constantly switch between multiple agents, Tutti is worth following.